2 * linux/drivers/block/ll_rw_blk.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
6 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
7 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
8 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> - July2000
9 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
13 * This handles all read/write requests to block devices
15 #include <linux/config.h>
16 #include <linux/kernel.h>
17 #include <linux/module.h>
18 #include <linux/backing-dev.h>
19 #include <linux/bio.h>
20 #include <linux/blkdev.h>
21 #include <linux/highmem.h>
23 #include <linux/kernel_stat.h>
24 #include <linux/string.h>
25 #include <linux/init.h>
26 #include <linux/bootmem.h> /* for max_pfn/max_low_pfn */
27 #include <linux/completion.h>
28 #include <linux/slab.h>
29 #include <linux/swap.h>
30 #include <linux/writeback.h>
31 #include <linux/blkdev.h>
36 #include <scsi/scsi_cmnd.h>
38 static void blk_unplug_work(void *data
);
39 static void blk_unplug_timeout(unsigned long data
);
40 static void drive_stat_acct(struct request
*rq
, int nr_sectors
, int new_io
);
43 * For the allocated request tables
45 static kmem_cache_t
*request_cachep
;
48 * For queue allocation
50 static kmem_cache_t
*requestq_cachep
;
53 * For io context allocations
55 static kmem_cache_t
*iocontext_cachep
;
57 static wait_queue_head_t congestion_wqh
[2] = {
58 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh
[0]),
59 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh
[1])
63 * Controlling structure to kblockd
65 static struct workqueue_struct
*kblockd_workqueue
;
67 unsigned long blk_max_low_pfn
, blk_max_pfn
;
69 EXPORT_SYMBOL(blk_max_low_pfn
);
70 EXPORT_SYMBOL(blk_max_pfn
);
72 /* Amount of time in which a process may batch requests */
73 #define BLK_BATCH_TIME (HZ/50UL)
75 /* Number of requests a "batching" process may submit */
76 #define BLK_BATCH_REQ 32
79 * Return the threshold (number of used requests) at which the queue is
80 * considered to be congested. It include a little hysteresis to keep the
81 * context switch rate down.
83 static inline int queue_congestion_on_threshold(struct request_queue
*q
)
85 return q
->nr_congestion_on
;
89 * The threshold at which a queue is considered to be uncongested
91 static inline int queue_congestion_off_threshold(struct request_queue
*q
)
93 return q
->nr_congestion_off
;
96 static void blk_queue_congestion_threshold(struct request_queue
*q
)
100 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
101 if (nr
> q
->nr_requests
)
103 q
->nr_congestion_on
= nr
;
105 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
108 q
->nr_congestion_off
= nr
;
112 * A queue has just exitted congestion. Note this in the global counter of
113 * congested queues, and wake up anyone who was waiting for requests to be
116 static void clear_queue_congested(request_queue_t
*q
, int rw
)
119 wait_queue_head_t
*wqh
= &congestion_wqh
[rw
];
121 bit
= (rw
== WRITE
) ? BDI_write_congested
: BDI_read_congested
;
122 clear_bit(bit
, &q
->backing_dev_info
.state
);
123 smp_mb__after_clear_bit();
124 if (waitqueue_active(wqh
))
129 * A queue has just entered congestion. Flag that in the queue's VM-visible
130 * state flags and increment the global gounter of congested queues.
132 static void set_queue_congested(request_queue_t
*q
, int rw
)
136 bit
= (rw
== WRITE
) ? BDI_write_congested
: BDI_read_congested
;
137 set_bit(bit
, &q
->backing_dev_info
.state
);
141 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
144 * Locates the passed device's request queue and returns the address of its
147 * Will return NULL if the request queue cannot be located.
149 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
151 struct backing_dev_info
*ret
= NULL
;
152 request_queue_t
*q
= bdev_get_queue(bdev
);
155 ret
= &q
->backing_dev_info
;
159 EXPORT_SYMBOL(blk_get_backing_dev_info
);
161 void blk_queue_activity_fn(request_queue_t
*q
, activity_fn
*fn
, void *data
)
164 q
->activity_data
= data
;
167 EXPORT_SYMBOL(blk_queue_activity_fn
);
170 * blk_queue_prep_rq - set a prepare_request function for queue
172 * @pfn: prepare_request function
174 * It's possible for a queue to register a prepare_request callback which
175 * is invoked before the request is handed to the request_fn. The goal of
176 * the function is to prepare a request for I/O, it can be used to build a
177 * cdb from the request data for instance.
180 void blk_queue_prep_rq(request_queue_t
*q
, prep_rq_fn
*pfn
)
185 EXPORT_SYMBOL(blk_queue_prep_rq
);
188 * blk_queue_merge_bvec - set a merge_bvec function for queue
190 * @mbfn: merge_bvec_fn
192 * Usually queues have static limitations on the max sectors or segments that
193 * we can put in a request. Stacking drivers may have some settings that
194 * are dynamic, and thus we have to query the queue whether it is ok to
195 * add a new bio_vec to a bio at a given offset or not. If the block device
196 * has such limitations, it needs to register a merge_bvec_fn to control
197 * the size of bio's sent to it. Note that a block device *must* allow a
198 * single page to be added to an empty bio. The block device driver may want
199 * to use the bio_split() function to deal with these bio's. By default
200 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
203 void blk_queue_merge_bvec(request_queue_t
*q
, merge_bvec_fn
*mbfn
)
205 q
->merge_bvec_fn
= mbfn
;
208 EXPORT_SYMBOL(blk_queue_merge_bvec
);
211 * blk_queue_make_request - define an alternate make_request function for a device
212 * @q: the request queue for the device to be affected
213 * @mfn: the alternate make_request function
216 * The normal way for &struct bios to be passed to a device
217 * driver is for them to be collected into requests on a request
218 * queue, and then to allow the device driver to select requests
219 * off that queue when it is ready. This works well for many block
220 * devices. However some block devices (typically virtual devices
221 * such as md or lvm) do not benefit from the processing on the
222 * request queue, and are served best by having the requests passed
223 * directly to them. This can be achieved by providing a function
224 * to blk_queue_make_request().
227 * The driver that does this *must* be able to deal appropriately
228 * with buffers in "highmemory". This can be accomplished by either calling
229 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
230 * blk_queue_bounce() to create a buffer in normal memory.
232 void blk_queue_make_request(request_queue_t
* q
, make_request_fn
* mfn
)
237 q
->nr_requests
= BLKDEV_MAX_RQ
;
238 blk_queue_max_phys_segments(q
, MAX_PHYS_SEGMENTS
);
239 blk_queue_max_hw_segments(q
, MAX_HW_SEGMENTS
);
240 q
->make_request_fn
= mfn
;
241 q
->backing_dev_info
.ra_pages
= (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
242 q
->backing_dev_info
.state
= 0;
243 q
->backing_dev_info
.capabilities
= BDI_CAP_MAP_COPY
;
244 blk_queue_max_sectors(q
, MAX_SECTORS
);
245 blk_queue_hardsect_size(q
, 512);
246 blk_queue_dma_alignment(q
, 511);
247 blk_queue_congestion_threshold(q
);
248 q
->nr_batching
= BLK_BATCH_REQ
;
250 q
->unplug_thresh
= 4; /* hmm */
251 q
->unplug_delay
= (3 * HZ
) / 1000; /* 3 milliseconds */
252 if (q
->unplug_delay
== 0)
255 INIT_WORK(&q
->unplug_work
, blk_unplug_work
, q
);
257 q
->unplug_timer
.function
= blk_unplug_timeout
;
258 q
->unplug_timer
.data
= (unsigned long)q
;
261 * by default assume old behaviour and bounce for any highmem page
263 blk_queue_bounce_limit(q
, BLK_BOUNCE_HIGH
);
265 blk_queue_activity_fn(q
, NULL
, NULL
);
267 INIT_LIST_HEAD(&q
->drain_list
);
270 EXPORT_SYMBOL(blk_queue_make_request
);
272 static inline void rq_init(request_queue_t
*q
, struct request
*rq
)
274 INIT_LIST_HEAD(&rq
->queuelist
);
277 rq
->rq_status
= RQ_ACTIVE
;
278 rq
->bio
= rq
->biotail
= NULL
;
287 rq
->nr_phys_segments
= 0;
290 rq
->end_io_data
= NULL
;
294 * blk_queue_ordered - does this queue support ordered writes
295 * @q: the request queue
299 * For journalled file systems, doing ordered writes on a commit
300 * block instead of explicitly doing wait_on_buffer (which is bad
301 * for performance) can be a big win. Block drivers supporting this
302 * feature should call this function and indicate so.
305 void blk_queue_ordered(request_queue_t
*q
, int flag
)
308 case QUEUE_ORDERED_NONE
:
310 kmem_cache_free(request_cachep
, q
->flush_rq
);
314 case QUEUE_ORDERED_TAG
:
317 case QUEUE_ORDERED_FLUSH
:
320 q
->flush_rq
= kmem_cache_alloc(request_cachep
,
324 printk("blk_queue_ordered: bad value %d\n", flag
);
329 EXPORT_SYMBOL(blk_queue_ordered
);
332 * blk_queue_issue_flush_fn - set function for issuing a flush
333 * @q: the request queue
334 * @iff: the function to be called issuing the flush
337 * If a driver supports issuing a flush command, the support is notified
338 * to the block layer by defining it through this call.
341 void blk_queue_issue_flush_fn(request_queue_t
*q
, issue_flush_fn
*iff
)
343 q
->issue_flush_fn
= iff
;
346 EXPORT_SYMBOL(blk_queue_issue_flush_fn
);
349 * Cache flushing for ordered writes handling
351 static void blk_pre_flush_end_io(struct request
*flush_rq
)
353 struct request
*rq
= flush_rq
->end_io_data
;
354 request_queue_t
*q
= rq
->q
;
356 elv_completed_request(q
, flush_rq
);
358 rq
->flags
|= REQ_BAR_PREFLUSH
;
360 if (!flush_rq
->errors
)
361 elv_requeue_request(q
, rq
);
363 q
->end_flush_fn(q
, flush_rq
);
364 clear_bit(QUEUE_FLAG_FLUSH
, &q
->queue_flags
);
369 static void blk_post_flush_end_io(struct request
*flush_rq
)
371 struct request
*rq
= flush_rq
->end_io_data
;
372 request_queue_t
*q
= rq
->q
;
374 elv_completed_request(q
, flush_rq
);
376 rq
->flags
|= REQ_BAR_POSTFLUSH
;
378 q
->end_flush_fn(q
, flush_rq
);
379 clear_bit(QUEUE_FLAG_FLUSH
, &q
->queue_flags
);
383 struct request
*blk_start_pre_flush(request_queue_t
*q
, struct request
*rq
)
385 struct request
*flush_rq
= q
->flush_rq
;
387 BUG_ON(!blk_barrier_rq(rq
));
389 if (test_and_set_bit(QUEUE_FLAG_FLUSH
, &q
->queue_flags
))
392 rq_init(q
, flush_rq
);
393 flush_rq
->elevator_private
= NULL
;
394 flush_rq
->flags
= REQ_BAR_FLUSH
;
395 flush_rq
->rq_disk
= rq
->rq_disk
;
399 * prepare_flush returns 0 if no flush is needed, just mark both
400 * pre and post flush as done in that case
402 if (!q
->prepare_flush_fn(q
, flush_rq
)) {
403 rq
->flags
|= REQ_BAR_PREFLUSH
| REQ_BAR_POSTFLUSH
;
404 clear_bit(QUEUE_FLAG_FLUSH
, &q
->queue_flags
);
409 * some drivers dequeue requests right away, some only after io
410 * completion. make sure the request is dequeued.
412 if (!list_empty(&rq
->queuelist
))
413 blkdev_dequeue_request(rq
);
415 flush_rq
->end_io_data
= rq
;
416 flush_rq
->end_io
= blk_pre_flush_end_io
;
418 __elv_add_request(q
, flush_rq
, ELEVATOR_INSERT_FRONT
, 0);
422 static void blk_start_post_flush(request_queue_t
*q
, struct request
*rq
)
424 struct request
*flush_rq
= q
->flush_rq
;
426 BUG_ON(!blk_barrier_rq(rq
));
428 rq_init(q
, flush_rq
);
429 flush_rq
->elevator_private
= NULL
;
430 flush_rq
->flags
= REQ_BAR_FLUSH
;
431 flush_rq
->rq_disk
= rq
->rq_disk
;
434 if (q
->prepare_flush_fn(q
, flush_rq
)) {
435 flush_rq
->end_io_data
= rq
;
436 flush_rq
->end_io
= blk_post_flush_end_io
;
438 __elv_add_request(q
, flush_rq
, ELEVATOR_INSERT_FRONT
, 0);
443 static inline int blk_check_end_barrier(request_queue_t
*q
, struct request
*rq
,
446 if (sectors
> rq
->nr_sectors
)
447 sectors
= rq
->nr_sectors
;
449 rq
->nr_sectors
-= sectors
;
450 return rq
->nr_sectors
;
453 static int __blk_complete_barrier_rq(request_queue_t
*q
, struct request
*rq
,
454 int sectors
, int queue_locked
)
456 if (q
->ordered
!= QUEUE_ORDERED_FLUSH
)
458 if (!blk_fs_request(rq
) || !blk_barrier_rq(rq
))
460 if (blk_barrier_postflush(rq
))
463 if (!blk_check_end_barrier(q
, rq
, sectors
)) {
464 unsigned long flags
= 0;
467 spin_lock_irqsave(q
->queue_lock
, flags
);
469 blk_start_post_flush(q
, rq
);
472 spin_unlock_irqrestore(q
->queue_lock
, flags
);
479 * blk_complete_barrier_rq - complete possible barrier request
480 * @q: the request queue for the device
482 * @sectors: number of sectors to complete
485 * Used in driver end_io handling to determine whether to postpone
486 * completion of a barrier request until a post flush has been done. This
487 * is the unlocked variant, used if the caller doesn't already hold the
490 int blk_complete_barrier_rq(request_queue_t
*q
, struct request
*rq
, int sectors
)
492 return __blk_complete_barrier_rq(q
, rq
, sectors
, 0);
494 EXPORT_SYMBOL(blk_complete_barrier_rq
);
497 * blk_complete_barrier_rq_locked - complete possible barrier request
498 * @q: the request queue for the device
500 * @sectors: number of sectors to complete
503 * See blk_complete_barrier_rq(). This variant must be used if the caller
504 * holds the queue lock.
506 int blk_complete_barrier_rq_locked(request_queue_t
*q
, struct request
*rq
,
509 return __blk_complete_barrier_rq(q
, rq
, sectors
, 1);
511 EXPORT_SYMBOL(blk_complete_barrier_rq_locked
);
514 * blk_queue_bounce_limit - set bounce buffer limit for queue
515 * @q: the request queue for the device
516 * @dma_addr: bus address limit
519 * Different hardware can have different requirements as to what pages
520 * it can do I/O directly to. A low level driver can call
521 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
522 * buffers for doing I/O to pages residing above @page. By default
523 * the block layer sets this to the highest numbered "low" memory page.
525 void blk_queue_bounce_limit(request_queue_t
*q
, u64 dma_addr
)
527 unsigned long bounce_pfn
= dma_addr
>> PAGE_SHIFT
;
530 * set appropriate bounce gfp mask -- unfortunately we don't have a
531 * full 4GB zone, so we have to resort to low memory for any bounces.
532 * ISA has its own < 16MB zone.
534 if (bounce_pfn
< blk_max_low_pfn
) {
535 BUG_ON(dma_addr
< BLK_BOUNCE_ISA
);
536 init_emergency_isa_pool();
537 q
->bounce_gfp
= GFP_NOIO
| GFP_DMA
;
539 q
->bounce_gfp
= GFP_NOIO
;
541 q
->bounce_pfn
= bounce_pfn
;
544 EXPORT_SYMBOL(blk_queue_bounce_limit
);
547 * blk_queue_max_sectors - set max sectors for a request for this queue
548 * @q: the request queue for the device
549 * @max_sectors: max sectors in the usual 512b unit
552 * Enables a low level driver to set an upper limit on the size of
555 void blk_queue_max_sectors(request_queue_t
*q
, unsigned short max_sectors
)
557 if ((max_sectors
<< 9) < PAGE_CACHE_SIZE
) {
558 max_sectors
= 1 << (PAGE_CACHE_SHIFT
- 9);
559 printk("%s: set to minimum %d\n", __FUNCTION__
, max_sectors
);
562 q
->max_sectors
= q
->max_hw_sectors
= max_sectors
;
565 EXPORT_SYMBOL(blk_queue_max_sectors
);
568 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
569 * @q: the request queue for the device
570 * @max_segments: max number of segments
573 * Enables a low level driver to set an upper limit on the number of
574 * physical data segments in a request. This would be the largest sized
575 * scatter list the driver could handle.
577 void blk_queue_max_phys_segments(request_queue_t
*q
, unsigned short max_segments
)
581 printk("%s: set to minimum %d\n", __FUNCTION__
, max_segments
);
584 q
->max_phys_segments
= max_segments
;
587 EXPORT_SYMBOL(blk_queue_max_phys_segments
);
590 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
591 * @q: the request queue for the device
592 * @max_segments: max number of segments
595 * Enables a low level driver to set an upper limit on the number of
596 * hw data segments in a request. This would be the largest number of
597 * address/length pairs the host adapter can actually give as once
600 void blk_queue_max_hw_segments(request_queue_t
*q
, unsigned short max_segments
)
604 printk("%s: set to minimum %d\n", __FUNCTION__
, max_segments
);
607 q
->max_hw_segments
= max_segments
;
610 EXPORT_SYMBOL(blk_queue_max_hw_segments
);
613 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
614 * @q: the request queue for the device
615 * @max_size: max size of segment in bytes
618 * Enables a low level driver to set an upper limit on the size of a
621 void blk_queue_max_segment_size(request_queue_t
*q
, unsigned int max_size
)
623 if (max_size
< PAGE_CACHE_SIZE
) {
624 max_size
= PAGE_CACHE_SIZE
;
625 printk("%s: set to minimum %d\n", __FUNCTION__
, max_size
);
628 q
->max_segment_size
= max_size
;
631 EXPORT_SYMBOL(blk_queue_max_segment_size
);
634 * blk_queue_hardsect_size - set hardware sector size for the queue
635 * @q: the request queue for the device
636 * @size: the hardware sector size, in bytes
639 * This should typically be set to the lowest possible sector size
640 * that the hardware can operate on (possible without reverting to
641 * even internal read-modify-write operations). Usually the default
642 * of 512 covers most hardware.
644 void blk_queue_hardsect_size(request_queue_t
*q
, unsigned short size
)
646 q
->hardsect_size
= size
;
649 EXPORT_SYMBOL(blk_queue_hardsect_size
);
652 * Returns the minimum that is _not_ zero, unless both are zero.
654 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
657 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
658 * @t: the stacking driver (top)
659 * @b: the underlying device (bottom)
661 void blk_queue_stack_limits(request_queue_t
*t
, request_queue_t
*b
)
663 /* zero is "infinity" */
664 t
->max_sectors
= t
->max_hw_sectors
=
665 min_not_zero(t
->max_sectors
,b
->max_sectors
);
667 t
->max_phys_segments
= min(t
->max_phys_segments
,b
->max_phys_segments
);
668 t
->max_hw_segments
= min(t
->max_hw_segments
,b
->max_hw_segments
);
669 t
->max_segment_size
= min(t
->max_segment_size
,b
->max_segment_size
);
670 t
->hardsect_size
= max(t
->hardsect_size
,b
->hardsect_size
);
673 EXPORT_SYMBOL(blk_queue_stack_limits
);
676 * blk_queue_segment_boundary - set boundary rules for segment merging
677 * @q: the request queue for the device
678 * @mask: the memory boundary mask
680 void blk_queue_segment_boundary(request_queue_t
*q
, unsigned long mask
)
682 if (mask
< PAGE_CACHE_SIZE
- 1) {
683 mask
= PAGE_CACHE_SIZE
- 1;
684 printk("%s: set to minimum %lx\n", __FUNCTION__
, mask
);
687 q
->seg_boundary_mask
= mask
;
690 EXPORT_SYMBOL(blk_queue_segment_boundary
);
693 * blk_queue_dma_alignment - set dma length and memory alignment
694 * @q: the request queue for the device
695 * @mask: alignment mask
698 * set required memory and length aligment for direct dma transactions.
699 * this is used when buiding direct io requests for the queue.
702 void blk_queue_dma_alignment(request_queue_t
*q
, int mask
)
704 q
->dma_alignment
= mask
;
707 EXPORT_SYMBOL(blk_queue_dma_alignment
);
710 * blk_queue_find_tag - find a request by its tag and queue
712 * @q: The request queue for the device
713 * @tag: The tag of the request
716 * Should be used when a device returns a tag and you want to match
719 * no locks need be held.
721 struct request
*blk_queue_find_tag(request_queue_t
*q
, int tag
)
723 struct blk_queue_tag
*bqt
= q
->queue_tags
;
725 if (unlikely(bqt
== NULL
|| tag
>= bqt
->real_max_depth
))
728 return bqt
->tag_index
[tag
];
731 EXPORT_SYMBOL(blk_queue_find_tag
);
734 * __blk_queue_free_tags - release tag maintenance info
735 * @q: the request queue for the device
738 * blk_cleanup_queue() will take care of calling this function, if tagging
739 * has been used. So there's no need to call this directly.
741 static void __blk_queue_free_tags(request_queue_t
*q
)
743 struct blk_queue_tag
*bqt
= q
->queue_tags
;
748 if (atomic_dec_and_test(&bqt
->refcnt
)) {
750 BUG_ON(!list_empty(&bqt
->busy_list
));
752 kfree(bqt
->tag_index
);
753 bqt
->tag_index
= NULL
;
761 q
->queue_tags
= NULL
;
762 q
->queue_flags
&= ~(1 << QUEUE_FLAG_QUEUED
);
766 * blk_queue_free_tags - release tag maintenance info
767 * @q: the request queue for the device
770 * This is used to disabled tagged queuing to a device, yet leave
773 void blk_queue_free_tags(request_queue_t
*q
)
775 clear_bit(QUEUE_FLAG_QUEUED
, &q
->queue_flags
);
778 EXPORT_SYMBOL(blk_queue_free_tags
);
781 init_tag_map(request_queue_t
*q
, struct blk_queue_tag
*tags
, int depth
)
783 struct request
**tag_index
;
784 unsigned long *tag_map
;
787 if (depth
> q
->nr_requests
* 2) {
788 depth
= q
->nr_requests
* 2;
789 printk(KERN_ERR
"%s: adjusted depth to %d\n",
790 __FUNCTION__
, depth
);
793 tag_index
= kmalloc(depth
* sizeof(struct request
*), GFP_ATOMIC
);
797 nr_ulongs
= ALIGN(depth
, BITS_PER_LONG
) / BITS_PER_LONG
;
798 tag_map
= kmalloc(nr_ulongs
* sizeof(unsigned long), GFP_ATOMIC
);
802 memset(tag_index
, 0, depth
* sizeof(struct request
*));
803 memset(tag_map
, 0, nr_ulongs
* sizeof(unsigned long));
804 tags
->real_max_depth
= depth
;
805 tags
->max_depth
= depth
;
806 tags
->tag_index
= tag_index
;
807 tags
->tag_map
= tag_map
;
816 * blk_queue_init_tags - initialize the queue tag info
817 * @q: the request queue for the device
818 * @depth: the maximum queue depth supported
819 * @tags: the tag to use
821 int blk_queue_init_tags(request_queue_t
*q
, int depth
,
822 struct blk_queue_tag
*tags
)
826 BUG_ON(tags
&& q
->queue_tags
&& tags
!= q
->queue_tags
);
828 if (!tags
&& !q
->queue_tags
) {
829 tags
= kmalloc(sizeof(struct blk_queue_tag
), GFP_ATOMIC
);
833 if (init_tag_map(q
, tags
, depth
))
836 INIT_LIST_HEAD(&tags
->busy_list
);
838 atomic_set(&tags
->refcnt
, 1);
839 } else if (q
->queue_tags
) {
840 if ((rc
= blk_queue_resize_tags(q
, depth
)))
842 set_bit(QUEUE_FLAG_QUEUED
, &q
->queue_flags
);
845 atomic_inc(&tags
->refcnt
);
848 * assign it, all done
850 q
->queue_tags
= tags
;
851 q
->queue_flags
|= (1 << QUEUE_FLAG_QUEUED
);
858 EXPORT_SYMBOL(blk_queue_init_tags
);
861 * blk_queue_resize_tags - change the queueing depth
862 * @q: the request queue for the device
863 * @new_depth: the new max command queueing depth
866 * Must be called with the queue lock held.
868 int blk_queue_resize_tags(request_queue_t
*q
, int new_depth
)
870 struct blk_queue_tag
*bqt
= q
->queue_tags
;
871 struct request
**tag_index
;
872 unsigned long *tag_map
;
873 int max_depth
, nr_ulongs
;
879 * if we already have large enough real_max_depth. just
880 * adjust max_depth. *NOTE* as requests with tag value
881 * between new_depth and real_max_depth can be in-flight, tag
882 * map can not be shrunk blindly here.
884 if (new_depth
<= bqt
->real_max_depth
) {
885 bqt
->max_depth
= new_depth
;
890 * save the old state info, so we can copy it back
892 tag_index
= bqt
->tag_index
;
893 tag_map
= bqt
->tag_map
;
894 max_depth
= bqt
->real_max_depth
;
896 if (init_tag_map(q
, bqt
, new_depth
))
899 memcpy(bqt
->tag_index
, tag_index
, max_depth
* sizeof(struct request
*));
900 nr_ulongs
= ALIGN(max_depth
, BITS_PER_LONG
) / BITS_PER_LONG
;
901 memcpy(bqt
->tag_map
, tag_map
, nr_ulongs
* sizeof(unsigned long));
908 EXPORT_SYMBOL(blk_queue_resize_tags
);
911 * blk_queue_end_tag - end tag operations for a request
912 * @q: the request queue for the device
913 * @rq: the request that has completed
916 * Typically called when end_that_request_first() returns 0, meaning
917 * all transfers have been done for a request. It's important to call
918 * this function before end_that_request_last(), as that will put the
919 * request back on the free list thus corrupting the internal tag list.
922 * queue lock must be held.
924 void blk_queue_end_tag(request_queue_t
*q
, struct request
*rq
)
926 struct blk_queue_tag
*bqt
= q
->queue_tags
;
931 if (unlikely(tag
>= bqt
->real_max_depth
))
933 * This can happen after tag depth has been reduced.
934 * FIXME: how about a warning or info message here?
938 if (unlikely(!__test_and_clear_bit(tag
, bqt
->tag_map
))) {
939 printk(KERN_ERR
"%s: attempt to clear non-busy tag (%d)\n",
944 list_del_init(&rq
->queuelist
);
945 rq
->flags
&= ~REQ_QUEUED
;
948 if (unlikely(bqt
->tag_index
[tag
] == NULL
))
949 printk(KERN_ERR
"%s: tag %d is missing\n",
952 bqt
->tag_index
[tag
] = NULL
;
956 EXPORT_SYMBOL(blk_queue_end_tag
);
959 * blk_queue_start_tag - find a free tag and assign it
960 * @q: the request queue for the device
961 * @rq: the block request that needs tagging
964 * This can either be used as a stand-alone helper, or possibly be
965 * assigned as the queue &prep_rq_fn (in which case &struct request
966 * automagically gets a tag assigned). Note that this function
967 * assumes that any type of request can be queued! if this is not
968 * true for your device, you must check the request type before
969 * calling this function. The request will also be removed from
970 * the request queue, so it's the drivers responsibility to readd
971 * it if it should need to be restarted for some reason.
974 * queue lock must be held.
976 int blk_queue_start_tag(request_queue_t
*q
, struct request
*rq
)
978 struct blk_queue_tag
*bqt
= q
->queue_tags
;
981 if (unlikely((rq
->flags
& REQ_QUEUED
))) {
983 "%s: request %p for device [%s] already tagged %d",
985 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->tag
);
989 tag
= find_first_zero_bit(bqt
->tag_map
, bqt
->max_depth
);
990 if (tag
>= bqt
->max_depth
)
993 __set_bit(tag
, bqt
->tag_map
);
995 rq
->flags
|= REQ_QUEUED
;
997 bqt
->tag_index
[tag
] = rq
;
998 blkdev_dequeue_request(rq
);
999 list_add(&rq
->queuelist
, &bqt
->busy_list
);
1004 EXPORT_SYMBOL(blk_queue_start_tag
);
1007 * blk_queue_invalidate_tags - invalidate all pending tags
1008 * @q: the request queue for the device
1011 * Hardware conditions may dictate a need to stop all pending requests.
1012 * In this case, we will safely clear the block side of the tag queue and
1013 * readd all requests to the request queue in the right order.
1016 * queue lock must be held.
1018 void blk_queue_invalidate_tags(request_queue_t
*q
)
1020 struct blk_queue_tag
*bqt
= q
->queue_tags
;
1021 struct list_head
*tmp
, *n
;
1024 list_for_each_safe(tmp
, n
, &bqt
->busy_list
) {
1025 rq
= list_entry_rq(tmp
);
1027 if (rq
->tag
== -1) {
1029 "%s: bad tag found on list\n", __FUNCTION__
);
1030 list_del_init(&rq
->queuelist
);
1031 rq
->flags
&= ~REQ_QUEUED
;
1033 blk_queue_end_tag(q
, rq
);
1035 rq
->flags
&= ~REQ_STARTED
;
1036 __elv_add_request(q
, rq
, ELEVATOR_INSERT_BACK
, 0);
1040 EXPORT_SYMBOL(blk_queue_invalidate_tags
);
1042 static char *rq_flags
[] = {
1061 "REQ_DRIVE_TASKFILE",
1068 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
1072 printk("%s: dev %s: flags = ", msg
,
1073 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?");
1076 if (rq
->flags
& (1 << bit
))
1077 printk("%s ", rq_flags
[bit
]);
1079 } while (bit
< __REQ_NR_BITS
);
1081 printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq
->sector
,
1083 rq
->current_nr_sectors
);
1084 printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq
->bio
, rq
->biotail
, rq
->buffer
, rq
->data
, rq
->data_len
);
1086 if (rq
->flags
& (REQ_BLOCK_PC
| REQ_PC
)) {
1088 for (bit
= 0; bit
< sizeof(rq
->cmd
); bit
++)
1089 printk("%02x ", rq
->cmd
[bit
]);
1094 EXPORT_SYMBOL(blk_dump_rq_flags
);
1096 void blk_recount_segments(request_queue_t
*q
, struct bio
*bio
)
1098 struct bio_vec
*bv
, *bvprv
= NULL
;
1099 int i
, nr_phys_segs
, nr_hw_segs
, seg_size
, hw_seg_size
, cluster
;
1100 int high
, highprv
= 1;
1102 if (unlikely(!bio
->bi_io_vec
))
1105 cluster
= q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
);
1106 hw_seg_size
= seg_size
= nr_phys_segs
= nr_hw_segs
= 0;
1107 bio_for_each_segment(bv
, bio
, i
) {
1109 * the trick here is making sure that a high page is never
1110 * considered part of another segment, since that might
1111 * change with the bounce page.
1113 high
= page_to_pfn(bv
->bv_page
) >= q
->bounce_pfn
;
1114 if (high
|| highprv
)
1115 goto new_hw_segment
;
1117 if (seg_size
+ bv
->bv_len
> q
->max_segment_size
)
1119 if (!BIOVEC_PHYS_MERGEABLE(bvprv
, bv
))
1121 if (!BIOVEC_SEG_BOUNDARY(q
, bvprv
, bv
))
1123 if (BIOVEC_VIRT_OVERSIZE(hw_seg_size
+ bv
->bv_len
))
1124 goto new_hw_segment
;
1126 seg_size
+= bv
->bv_len
;
1127 hw_seg_size
+= bv
->bv_len
;
1132 if (BIOVEC_VIRT_MERGEABLE(bvprv
, bv
) &&
1133 !BIOVEC_VIRT_OVERSIZE(hw_seg_size
+ bv
->bv_len
)) {
1134 hw_seg_size
+= bv
->bv_len
;
1137 if (hw_seg_size
> bio
->bi_hw_front_size
)
1138 bio
->bi_hw_front_size
= hw_seg_size
;
1139 hw_seg_size
= BIOVEC_VIRT_START_SIZE(bv
) + bv
->bv_len
;
1145 seg_size
= bv
->bv_len
;
1148 if (hw_seg_size
> bio
->bi_hw_back_size
)
1149 bio
->bi_hw_back_size
= hw_seg_size
;
1150 if (nr_hw_segs
== 1 && hw_seg_size
> bio
->bi_hw_front_size
)
1151 bio
->bi_hw_front_size
= hw_seg_size
;
1152 bio
->bi_phys_segments
= nr_phys_segs
;
1153 bio
->bi_hw_segments
= nr_hw_segs
;
1154 bio
->bi_flags
|= (1 << BIO_SEG_VALID
);
1158 static int blk_phys_contig_segment(request_queue_t
*q
, struct bio
*bio
,
1161 if (!(q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
)))
1164 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio
), __BVEC_START(nxt
)))
1166 if (bio
->bi_size
+ nxt
->bi_size
> q
->max_segment_size
)
1170 * bio and nxt are contigous in memory, check if the queue allows
1171 * these two to be merged into one
1173 if (BIO_SEG_BOUNDARY(q
, bio
, nxt
))
1179 static int blk_hw_contig_segment(request_queue_t
*q
, struct bio
*bio
,
1182 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1183 blk_recount_segments(q
, bio
);
1184 if (unlikely(!bio_flagged(nxt
, BIO_SEG_VALID
)))
1185 blk_recount_segments(q
, nxt
);
1186 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio
), __BVEC_START(nxt
)) ||
1187 BIOVEC_VIRT_OVERSIZE(bio
->bi_hw_front_size
+ bio
->bi_hw_back_size
))
1189 if (bio
->bi_size
+ nxt
->bi_size
> q
->max_segment_size
)
1196 * map a request to scatterlist, return number of sg entries setup. Caller
1197 * must make sure sg can hold rq->nr_phys_segments entries
1199 int blk_rq_map_sg(request_queue_t
*q
, struct request
*rq
, struct scatterlist
*sg
)
1201 struct bio_vec
*bvec
, *bvprv
;
1203 int nsegs
, i
, cluster
;
1206 cluster
= q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
);
1209 * for each bio in rq
1212 rq_for_each_bio(bio
, rq
) {
1214 * for each segment in bio
1216 bio_for_each_segment(bvec
, bio
, i
) {
1217 int nbytes
= bvec
->bv_len
;
1219 if (bvprv
&& cluster
) {
1220 if (sg
[nsegs
- 1].length
+ nbytes
> q
->max_segment_size
)
1223 if (!BIOVEC_PHYS_MERGEABLE(bvprv
, bvec
))
1225 if (!BIOVEC_SEG_BOUNDARY(q
, bvprv
, bvec
))
1228 sg
[nsegs
- 1].length
+= nbytes
;
1231 memset(&sg
[nsegs
],0,sizeof(struct scatterlist
));
1232 sg
[nsegs
].page
= bvec
->bv_page
;
1233 sg
[nsegs
].length
= nbytes
;
1234 sg
[nsegs
].offset
= bvec
->bv_offset
;
1239 } /* segments in bio */
1245 EXPORT_SYMBOL(blk_rq_map_sg
);
1248 * the standard queue merge functions, can be overridden with device
1249 * specific ones if so desired
1252 static inline int ll_new_mergeable(request_queue_t
*q
,
1253 struct request
*req
,
1256 int nr_phys_segs
= bio_phys_segments(q
, bio
);
1258 if (req
->nr_phys_segments
+ nr_phys_segs
> q
->max_phys_segments
) {
1259 req
->flags
|= REQ_NOMERGE
;
1260 if (req
== q
->last_merge
)
1261 q
->last_merge
= NULL
;
1266 * A hw segment is just getting larger, bump just the phys
1269 req
->nr_phys_segments
+= nr_phys_segs
;
1273 static inline int ll_new_hw_segment(request_queue_t
*q
,
1274 struct request
*req
,
1277 int nr_hw_segs
= bio_hw_segments(q
, bio
);
1278 int nr_phys_segs
= bio_phys_segments(q
, bio
);
1280 if (req
->nr_hw_segments
+ nr_hw_segs
> q
->max_hw_segments
1281 || req
->nr_phys_segments
+ nr_phys_segs
> q
->max_phys_segments
) {
1282 req
->flags
|= REQ_NOMERGE
;
1283 if (req
== q
->last_merge
)
1284 q
->last_merge
= NULL
;
1289 * This will form the start of a new hw segment. Bump both
1292 req
->nr_hw_segments
+= nr_hw_segs
;
1293 req
->nr_phys_segments
+= nr_phys_segs
;
1297 static int ll_back_merge_fn(request_queue_t
*q
, struct request
*req
,
1302 if (req
->nr_sectors
+ bio_sectors(bio
) > q
->max_sectors
) {
1303 req
->flags
|= REQ_NOMERGE
;
1304 if (req
== q
->last_merge
)
1305 q
->last_merge
= NULL
;
1308 if (unlikely(!bio_flagged(req
->biotail
, BIO_SEG_VALID
)))
1309 blk_recount_segments(q
, req
->biotail
);
1310 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1311 blk_recount_segments(q
, bio
);
1312 len
= req
->biotail
->bi_hw_back_size
+ bio
->bi_hw_front_size
;
1313 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req
->biotail
), __BVEC_START(bio
)) &&
1314 !BIOVEC_VIRT_OVERSIZE(len
)) {
1315 int mergeable
= ll_new_mergeable(q
, req
, bio
);
1318 if (req
->nr_hw_segments
== 1)
1319 req
->bio
->bi_hw_front_size
= len
;
1320 if (bio
->bi_hw_segments
== 1)
1321 bio
->bi_hw_back_size
= len
;
1326 return ll_new_hw_segment(q
, req
, bio
);
1329 static int ll_front_merge_fn(request_queue_t
*q
, struct request
*req
,
1334 if (req
->nr_sectors
+ bio_sectors(bio
) > q
->max_sectors
) {
1335 req
->flags
|= REQ_NOMERGE
;
1336 if (req
== q
->last_merge
)
1337 q
->last_merge
= NULL
;
1340 len
= bio
->bi_hw_back_size
+ req
->bio
->bi_hw_front_size
;
1341 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1342 blk_recount_segments(q
, bio
);
1343 if (unlikely(!bio_flagged(req
->bio
, BIO_SEG_VALID
)))
1344 blk_recount_segments(q
, req
->bio
);
1345 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio
), __BVEC_START(req
->bio
)) &&
1346 !BIOVEC_VIRT_OVERSIZE(len
)) {
1347 int mergeable
= ll_new_mergeable(q
, req
, bio
);
1350 if (bio
->bi_hw_segments
== 1)
1351 bio
->bi_hw_front_size
= len
;
1352 if (req
->nr_hw_segments
== 1)
1353 req
->biotail
->bi_hw_back_size
= len
;
1358 return ll_new_hw_segment(q
, req
, bio
);
1361 static int ll_merge_requests_fn(request_queue_t
*q
, struct request
*req
,
1362 struct request
*next
)
1364 int total_phys_segments
;
1365 int total_hw_segments
;
1368 * First check if the either of the requests are re-queued
1369 * requests. Can't merge them if they are.
1371 if (req
->special
|| next
->special
)
1375 * Will it become too large?
1377 if ((req
->nr_sectors
+ next
->nr_sectors
) > q
->max_sectors
)
1380 total_phys_segments
= req
->nr_phys_segments
+ next
->nr_phys_segments
;
1381 if (blk_phys_contig_segment(q
, req
->biotail
, next
->bio
))
1382 total_phys_segments
--;
1384 if (total_phys_segments
> q
->max_phys_segments
)
1387 total_hw_segments
= req
->nr_hw_segments
+ next
->nr_hw_segments
;
1388 if (blk_hw_contig_segment(q
, req
->biotail
, next
->bio
)) {
1389 int len
= req
->biotail
->bi_hw_back_size
+ next
->bio
->bi_hw_front_size
;
1391 * propagate the combined length to the end of the requests
1393 if (req
->nr_hw_segments
== 1)
1394 req
->bio
->bi_hw_front_size
= len
;
1395 if (next
->nr_hw_segments
== 1)
1396 next
->biotail
->bi_hw_back_size
= len
;
1397 total_hw_segments
--;
1400 if (total_hw_segments
> q
->max_hw_segments
)
1403 /* Merge is OK... */
1404 req
->nr_phys_segments
= total_phys_segments
;
1405 req
->nr_hw_segments
= total_hw_segments
;
1410 * "plug" the device if there are no outstanding requests: this will
1411 * force the transfer to start only after we have put all the requests
1414 * This is called with interrupts off and no requests on the queue and
1415 * with the queue lock held.
1417 void blk_plug_device(request_queue_t
*q
)
1419 WARN_ON(!irqs_disabled());
1422 * don't plug a stopped queue, it must be paired with blk_start_queue()
1423 * which will restart the queueing
1425 if (test_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
))
1428 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
))
1429 mod_timer(&q
->unplug_timer
, jiffies
+ q
->unplug_delay
);
1432 EXPORT_SYMBOL(blk_plug_device
);
1435 * remove the queue from the plugged list, if present. called with
1436 * queue lock held and interrupts disabled.
1438 int blk_remove_plug(request_queue_t
*q
)
1440 WARN_ON(!irqs_disabled());
1442 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
))
1445 del_timer(&q
->unplug_timer
);
1449 EXPORT_SYMBOL(blk_remove_plug
);
1452 * remove the plug and let it rip..
1454 void __generic_unplug_device(request_queue_t
*q
)
1456 if (unlikely(test_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
)))
1459 if (!blk_remove_plug(q
))
1464 EXPORT_SYMBOL(__generic_unplug_device
);
1467 * generic_unplug_device - fire a request queue
1468 * @q: The &request_queue_t in question
1471 * Linux uses plugging to build bigger requests queues before letting
1472 * the device have at them. If a queue is plugged, the I/O scheduler
1473 * is still adding and merging requests on the queue. Once the queue
1474 * gets unplugged, the request_fn defined for the queue is invoked and
1475 * transfers started.
1477 void generic_unplug_device(request_queue_t
*q
)
1479 spin_lock_irq(q
->queue_lock
);
1480 __generic_unplug_device(q
);
1481 spin_unlock_irq(q
->queue_lock
);
1483 EXPORT_SYMBOL(generic_unplug_device
);
1485 static void blk_backing_dev_unplug(struct backing_dev_info
*bdi
,
1488 request_queue_t
*q
= bdi
->unplug_io_data
;
1491 * devices don't necessarily have an ->unplug_fn defined
1497 static void blk_unplug_work(void *data
)
1499 request_queue_t
*q
= data
;
1504 static void blk_unplug_timeout(unsigned long data
)
1506 request_queue_t
*q
= (request_queue_t
*)data
;
1508 kblockd_schedule_work(&q
->unplug_work
);
1512 * blk_start_queue - restart a previously stopped queue
1513 * @q: The &request_queue_t in question
1516 * blk_start_queue() will clear the stop flag on the queue, and call
1517 * the request_fn for the queue if it was in a stopped state when
1518 * entered. Also see blk_stop_queue(). Queue lock must be held.
1520 void blk_start_queue(request_queue_t
*q
)
1522 clear_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
);
1525 * one level of recursion is ok and is much faster than kicking
1526 * the unplug handling
1528 if (!test_and_set_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
)) {
1530 clear_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
);
1533 kblockd_schedule_work(&q
->unplug_work
);
1537 EXPORT_SYMBOL(blk_start_queue
);
1540 * blk_stop_queue - stop a queue
1541 * @q: The &request_queue_t in question
1544 * The Linux block layer assumes that a block driver will consume all
1545 * entries on the request queue when the request_fn strategy is called.
1546 * Often this will not happen, because of hardware limitations (queue
1547 * depth settings). If a device driver gets a 'queue full' response,
1548 * or if it simply chooses not to queue more I/O at one point, it can
1549 * call this function to prevent the request_fn from being called until
1550 * the driver has signalled it's ready to go again. This happens by calling
1551 * blk_start_queue() to restart queue operations. Queue lock must be held.
1553 void blk_stop_queue(request_queue_t
*q
)
1556 set_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
);
1558 EXPORT_SYMBOL(blk_stop_queue
);
1561 * blk_sync_queue - cancel any pending callbacks on a queue
1565 * The block layer may perform asynchronous callback activity
1566 * on a queue, such as calling the unplug function after a timeout.
1567 * A block device may call blk_sync_queue to ensure that any
1568 * such activity is cancelled, thus allowing it to release resources
1569 * the the callbacks might use. The caller must already have made sure
1570 * that its ->make_request_fn will not re-add plugging prior to calling
1574 void blk_sync_queue(struct request_queue
*q
)
1576 del_timer_sync(&q
->unplug_timer
);
1579 EXPORT_SYMBOL(blk_sync_queue
);
1582 * blk_run_queue - run a single device queue
1583 * @q: The queue to run
1585 void blk_run_queue(struct request_queue
*q
)
1587 unsigned long flags
;
1589 spin_lock_irqsave(q
->queue_lock
, flags
);
1591 if (!elv_queue_empty(q
))
1593 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1595 EXPORT_SYMBOL(blk_run_queue
);
1598 * blk_cleanup_queue: - release a &request_queue_t when it is no longer needed
1599 * @q: the request queue to be released
1602 * blk_cleanup_queue is the pair to blk_init_queue() or
1603 * blk_queue_make_request(). It should be called when a request queue is
1604 * being released; typically when a block device is being de-registered.
1605 * Currently, its primary task it to free all the &struct request
1606 * structures that were allocated to the queue and the queue itself.
1609 * Hopefully the low level driver will have finished any
1610 * outstanding requests first...
1612 void blk_cleanup_queue(request_queue_t
* q
)
1614 struct request_list
*rl
= &q
->rq
;
1616 if (!atomic_dec_and_test(&q
->refcnt
))
1620 elevator_exit(q
->elevator
);
1625 mempool_destroy(rl
->rq_pool
);
1628 __blk_queue_free_tags(q
);
1630 blk_queue_ordered(q
, QUEUE_ORDERED_NONE
);
1632 kmem_cache_free(requestq_cachep
, q
);
1635 EXPORT_SYMBOL(blk_cleanup_queue
);
1637 static int blk_init_free_list(request_queue_t
*q
)
1639 struct request_list
*rl
= &q
->rq
;
1641 rl
->count
[READ
] = rl
->count
[WRITE
] = 0;
1642 rl
->starved
[READ
] = rl
->starved
[WRITE
] = 0;
1643 init_waitqueue_head(&rl
->wait
[READ
]);
1644 init_waitqueue_head(&rl
->wait
[WRITE
]);
1645 init_waitqueue_head(&rl
->drain
);
1647 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
1648 mempool_free_slab
, request_cachep
, q
->node
);
1656 static int __make_request(request_queue_t
*, struct bio
*);
1658 request_queue_t
*blk_alloc_queue(int gfp_mask
)
1660 return blk_alloc_queue_node(gfp_mask
, -1);
1662 EXPORT_SYMBOL(blk_alloc_queue
);
1664 request_queue_t
*blk_alloc_queue_node(int gfp_mask
, int node_id
)
1668 q
= kmem_cache_alloc_node(requestq_cachep
, gfp_mask
, node_id
);
1672 memset(q
, 0, sizeof(*q
));
1673 init_timer(&q
->unplug_timer
);
1674 atomic_set(&q
->refcnt
, 1);
1676 q
->backing_dev_info
.unplug_io_fn
= blk_backing_dev_unplug
;
1677 q
->backing_dev_info
.unplug_io_data
= q
;
1681 EXPORT_SYMBOL(blk_alloc_queue_node
);
1684 * blk_init_queue - prepare a request queue for use with a block device
1685 * @rfn: The function to be called to process requests that have been
1686 * placed on the queue.
1687 * @lock: Request queue spin lock
1690 * If a block device wishes to use the standard request handling procedures,
1691 * which sorts requests and coalesces adjacent requests, then it must
1692 * call blk_init_queue(). The function @rfn will be called when there
1693 * are requests on the queue that need to be processed. If the device
1694 * supports plugging, then @rfn may not be called immediately when requests
1695 * are available on the queue, but may be called at some time later instead.
1696 * Plugged queues are generally unplugged when a buffer belonging to one
1697 * of the requests on the queue is needed, or due to memory pressure.
1699 * @rfn is not required, or even expected, to remove all requests off the
1700 * queue, but only as many as it can handle at a time. If it does leave
1701 * requests on the queue, it is responsible for arranging that the requests
1702 * get dealt with eventually.
1704 * The queue spin lock must be held while manipulating the requests on the
1707 * Function returns a pointer to the initialized request queue, or NULL if
1708 * it didn't succeed.
1711 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1712 * when the block device is deactivated (such as at module unload).
1715 request_queue_t
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
1717 return blk_init_queue_node(rfn
, lock
, -1);
1719 EXPORT_SYMBOL(blk_init_queue
);
1722 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
1724 request_queue_t
*q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
1730 if (blk_init_free_list(q
))
1734 * if caller didn't supply a lock, they get per-queue locking with
1738 spin_lock_init(&q
->__queue_lock
);
1739 lock
= &q
->__queue_lock
;
1742 q
->request_fn
= rfn
;
1743 q
->back_merge_fn
= ll_back_merge_fn
;
1744 q
->front_merge_fn
= ll_front_merge_fn
;
1745 q
->merge_requests_fn
= ll_merge_requests_fn
;
1746 q
->prep_rq_fn
= NULL
;
1747 q
->unplug_fn
= generic_unplug_device
;
1748 q
->queue_flags
= (1 << QUEUE_FLAG_CLUSTER
);
1749 q
->queue_lock
= lock
;
1751 blk_queue_segment_boundary(q
, 0xffffffff);
1753 blk_queue_make_request(q
, __make_request
);
1754 blk_queue_max_segment_size(q
, MAX_SEGMENT_SIZE
);
1756 blk_queue_max_hw_segments(q
, MAX_HW_SEGMENTS
);
1757 blk_queue_max_phys_segments(q
, MAX_PHYS_SEGMENTS
);
1762 if (!elevator_init(q
, NULL
)) {
1763 blk_queue_congestion_threshold(q
);
1767 blk_cleanup_queue(q
);
1769 kmem_cache_free(requestq_cachep
, q
);
1772 EXPORT_SYMBOL(blk_init_queue_node
);
1774 int blk_get_queue(request_queue_t
*q
)
1776 if (likely(!test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
))) {
1777 atomic_inc(&q
->refcnt
);
1784 EXPORT_SYMBOL(blk_get_queue
);
1786 static inline void blk_free_request(request_queue_t
*q
, struct request
*rq
)
1788 elv_put_request(q
, rq
);
1789 mempool_free(rq
, q
->rq
.rq_pool
);
1792 static inline struct request
*
1793 blk_alloc_request(request_queue_t
*q
, int rw
, struct bio
*bio
, int gfp_mask
)
1795 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
1801 * first three bits are identical in rq->flags and bio->bi_rw,
1802 * see bio.h and blkdev.h
1806 if (!elv_set_request(q
, rq
, bio
, gfp_mask
))
1809 mempool_free(rq
, q
->rq
.rq_pool
);
1814 * ioc_batching returns true if the ioc is a valid batching request and
1815 * should be given priority access to a request.
1817 static inline int ioc_batching(request_queue_t
*q
, struct io_context
*ioc
)
1823 * Make sure the process is able to allocate at least 1 request
1824 * even if the batch times out, otherwise we could theoretically
1827 return ioc
->nr_batch_requests
== q
->nr_batching
||
1828 (ioc
->nr_batch_requests
> 0
1829 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
1833 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1834 * will cause the process to be a "batcher" on all queues in the system. This
1835 * is the behaviour we want though - once it gets a wakeup it should be given
1838 static void ioc_set_batching(request_queue_t
*q
, struct io_context
*ioc
)
1840 if (!ioc
|| ioc_batching(q
, ioc
))
1843 ioc
->nr_batch_requests
= q
->nr_batching
;
1844 ioc
->last_waited
= jiffies
;
1847 static void __freed_request(request_queue_t
*q
, int rw
)
1849 struct request_list
*rl
= &q
->rq
;
1851 if (rl
->count
[rw
] < queue_congestion_off_threshold(q
))
1852 clear_queue_congested(q
, rw
);
1854 if (rl
->count
[rw
] + 1 <= q
->nr_requests
) {
1855 if (waitqueue_active(&rl
->wait
[rw
]))
1856 wake_up(&rl
->wait
[rw
]);
1858 blk_clear_queue_full(q
, rw
);
1863 * A request has just been released. Account for it, update the full and
1864 * congestion status, wake up any waiters. Called under q->queue_lock.
1866 static void freed_request(request_queue_t
*q
, int rw
)
1868 struct request_list
*rl
= &q
->rq
;
1872 __freed_request(q
, rw
);
1874 if (unlikely(rl
->starved
[rw
^ 1]))
1875 __freed_request(q
, rw
^ 1);
1877 if (!rl
->count
[READ
] && !rl
->count
[WRITE
]) {
1879 if (unlikely(waitqueue_active(&rl
->drain
)))
1880 wake_up(&rl
->drain
);
1884 #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
1886 * Get a free request, queue_lock must be held.
1887 * Returns NULL on failure, with queue_lock held.
1888 * Returns !NULL on success, with queue_lock *not held*.
1890 static struct request
*get_request(request_queue_t
*q
, int rw
, struct bio
*bio
,
1893 struct request
*rq
= NULL
;
1894 struct request_list
*rl
= &q
->rq
;
1895 struct io_context
*ioc
= current_io_context(GFP_ATOMIC
);
1897 if (unlikely(test_bit(QUEUE_FLAG_DRAIN
, &q
->queue_flags
)))
1900 if (rl
->count
[rw
]+1 >= q
->nr_requests
) {
1902 * The queue will fill after this allocation, so set it as
1903 * full, and mark this process as "batching". This process
1904 * will be allowed to complete a batch of requests, others
1907 if (!blk_queue_full(q
, rw
)) {
1908 ioc_set_batching(q
, ioc
);
1909 blk_set_queue_full(q
, rw
);
1913 switch (elv_may_queue(q
, rw
, bio
)) {
1916 case ELV_MQUEUE_MAY
:
1918 case ELV_MQUEUE_MUST
:
1922 if (blk_queue_full(q
, rw
) && !ioc_batching(q
, ioc
)) {
1924 * The queue is full and the allocating process is not a
1925 * "batcher", and not exempted by the IO scheduler
1932 * Only allow batching queuers to allocate up to 50% over the defined
1933 * limit of requests, otherwise we could have thousands of requests
1934 * allocated with any setting of ->nr_requests
1936 if (rl
->count
[rw
] >= (3 * q
->nr_requests
/ 2))
1940 rl
->starved
[rw
] = 0;
1941 if (rl
->count
[rw
] >= queue_congestion_on_threshold(q
))
1942 set_queue_congested(q
, rw
);
1943 spin_unlock_irq(q
->queue_lock
);
1945 rq
= blk_alloc_request(q
, rw
, bio
, gfp_mask
);
1948 * Allocation failed presumably due to memory. Undo anything
1949 * we might have messed up.
1951 * Allocating task should really be put onto the front of the
1952 * wait queue, but this is pretty rare.
1954 spin_lock_irq(q
->queue_lock
);
1955 freed_request(q
, rw
);
1958 * in the very unlikely event that allocation failed and no
1959 * requests for this direction was pending, mark us starved
1960 * so that freeing of a request in the other direction will
1961 * notice us. another possible fix would be to split the
1962 * rq mempool into READ and WRITE
1965 if (unlikely(rl
->count
[rw
] == 0))
1966 rl
->starved
[rw
] = 1;
1971 if (ioc_batching(q
, ioc
))
1972 ioc
->nr_batch_requests
--;
1981 * No available requests for this queue, unplug the device and wait for some
1982 * requests to become available.
1984 * Called with q->queue_lock held, and returns with it unlocked.
1986 static struct request
*get_request_wait(request_queue_t
*q
, int rw
,
1991 rq
= get_request(q
, rw
, bio
, GFP_NOIO
);
1994 struct request_list
*rl
= &q
->rq
;
1996 prepare_to_wait_exclusive(&rl
->wait
[rw
], &wait
,
1997 TASK_UNINTERRUPTIBLE
);
1999 rq
= get_request(q
, rw
, bio
, GFP_NOIO
);
2002 struct io_context
*ioc
;
2004 __generic_unplug_device(q
);
2005 spin_unlock_irq(q
->queue_lock
);
2009 * After sleeping, we become a "batching" process and
2010 * will be able to allocate at least one request, and
2011 * up to a big batch of them for a small period time.
2012 * See ioc_batching, ioc_set_batching
2014 ioc
= current_io_context(GFP_NOIO
);
2015 ioc_set_batching(q
, ioc
);
2017 spin_lock_irq(q
->queue_lock
);
2019 finish_wait(&rl
->wait
[rw
], &wait
);
2025 struct request
*blk_get_request(request_queue_t
*q
, int rw
, int gfp_mask
)
2029 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
2031 spin_lock_irq(q
->queue_lock
);
2032 if (gfp_mask
& __GFP_WAIT
) {
2033 rq
= get_request_wait(q
, rw
, NULL
);
2035 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
2037 spin_unlock_irq(q
->queue_lock
);
2039 /* q->queue_lock is unlocked at this point */
2043 EXPORT_SYMBOL(blk_get_request
);
2046 * blk_requeue_request - put a request back on queue
2047 * @q: request queue where request should be inserted
2048 * @rq: request to be inserted
2051 * Drivers often keep queueing requests until the hardware cannot accept
2052 * more, when that condition happens we need to put the request back
2053 * on the queue. Must be called with queue lock held.
2055 void blk_requeue_request(request_queue_t
*q
, struct request
*rq
)
2057 if (blk_rq_tagged(rq
))
2058 blk_queue_end_tag(q
, rq
);
2060 elv_requeue_request(q
, rq
);
2063 EXPORT_SYMBOL(blk_requeue_request
);
2066 * blk_insert_request - insert a special request in to a request queue
2067 * @q: request queue where request should be inserted
2068 * @rq: request to be inserted
2069 * @at_head: insert request at head or tail of queue
2070 * @data: private data
2073 * Many block devices need to execute commands asynchronously, so they don't
2074 * block the whole kernel from preemption during request execution. This is
2075 * accomplished normally by inserting aritficial requests tagged as
2076 * REQ_SPECIAL in to the corresponding request queue, and letting them be
2077 * scheduled for actual execution by the request queue.
2079 * We have the option of inserting the head or the tail of the queue.
2080 * Typically we use the tail for new ioctls and so forth. We use the head
2081 * of the queue for things like a QUEUE_FULL message from a device, or a
2082 * host that is unable to accept a particular command.
2084 void blk_insert_request(request_queue_t
*q
, struct request
*rq
,
2085 int at_head
, void *data
)
2087 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
2088 unsigned long flags
;
2091 * tell I/O scheduler that this isn't a regular read/write (ie it
2092 * must not attempt merges on this) and that it acts as a soft
2095 rq
->flags
|= REQ_SPECIAL
| REQ_SOFTBARRIER
;
2099 spin_lock_irqsave(q
->queue_lock
, flags
);
2102 * If command is tagged, release the tag
2104 if (blk_rq_tagged(rq
))
2105 blk_queue_end_tag(q
, rq
);
2107 drive_stat_acct(rq
, rq
->nr_sectors
, 1);
2108 __elv_add_request(q
, rq
, where
, 0);
2110 if (blk_queue_plugged(q
))
2111 __generic_unplug_device(q
);
2114 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2117 EXPORT_SYMBOL(blk_insert_request
);
2120 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
2121 * @q: request queue where request should be inserted
2122 * @rq: request structure to fill
2123 * @ubuf: the user buffer
2124 * @len: length of user data
2127 * Data will be mapped directly for zero copy io, if possible. Otherwise
2128 * a kernel bounce buffer is used.
2130 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2131 * still in process context.
2133 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2134 * before being submitted to the device, as pages mapped may be out of
2135 * reach. It's the callers responsibility to make sure this happens. The
2136 * original bio must be passed back in to blk_rq_unmap_user() for proper
2139 int blk_rq_map_user(request_queue_t
*q
, struct request
*rq
, void __user
*ubuf
,
2142 unsigned long uaddr
;
2146 if (len
> (q
->max_sectors
<< 9))
2151 reading
= rq_data_dir(rq
) == READ
;
2154 * if alignment requirement is satisfied, map in user pages for
2155 * direct dma. else, set up kernel bounce buffers
2157 uaddr
= (unsigned long) ubuf
;
2158 if (!(uaddr
& queue_dma_alignment(q
)) && !(len
& queue_dma_alignment(q
)))
2159 bio
= bio_map_user(q
, NULL
, uaddr
, len
, reading
);
2161 bio
= bio_copy_user(q
, uaddr
, len
, reading
);
2164 rq
->bio
= rq
->biotail
= bio
;
2165 blk_rq_bio_prep(q
, rq
, bio
);
2167 rq
->buffer
= rq
->data
= NULL
;
2173 * bio is the err-ptr
2175 return PTR_ERR(bio
);
2178 EXPORT_SYMBOL(blk_rq_map_user
);
2181 * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage
2182 * @q: request queue where request should be inserted
2183 * @rq: request to map data to
2184 * @iov: pointer to the iovec
2185 * @iov_count: number of elements in the iovec
2188 * Data will be mapped directly for zero copy io, if possible. Otherwise
2189 * a kernel bounce buffer is used.
2191 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2192 * still in process context.
2194 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2195 * before being submitted to the device, as pages mapped may be out of
2196 * reach. It's the callers responsibility to make sure this happens. The
2197 * original bio must be passed back in to blk_rq_unmap_user() for proper
2200 int blk_rq_map_user_iov(request_queue_t
*q
, struct request
*rq
,
2201 struct sg_iovec
*iov
, int iov_count
)
2205 if (!iov
|| iov_count
<= 0)
2208 /* we don't allow misaligned data like bio_map_user() does. If the
2209 * user is using sg, they're expected to know the alignment constraints
2210 * and respect them accordingly */
2211 bio
= bio_map_user_iov(q
, NULL
, iov
, iov_count
, rq_data_dir(rq
)== READ
);
2213 return PTR_ERR(bio
);
2215 rq
->bio
= rq
->biotail
= bio
;
2216 blk_rq_bio_prep(q
, rq
, bio
);
2217 rq
->buffer
= rq
->data
= NULL
;
2218 rq
->data_len
= bio
->bi_size
;
2222 EXPORT_SYMBOL(blk_rq_map_user_iov
);
2225 * blk_rq_unmap_user - unmap a request with user data
2226 * @bio: bio to be unmapped
2227 * @ulen: length of user buffer
2230 * Unmap a bio previously mapped by blk_rq_map_user().
2232 int blk_rq_unmap_user(struct bio
*bio
, unsigned int ulen
)
2237 if (bio_flagged(bio
, BIO_USER_MAPPED
))
2238 bio_unmap_user(bio
);
2240 ret
= bio_uncopy_user(bio
);
2246 EXPORT_SYMBOL(blk_rq_unmap_user
);
2249 * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage
2250 * @q: request queue where request should be inserted
2251 * @rq: request to fill
2252 * @kbuf: the kernel buffer
2253 * @len: length of user data
2254 * @gfp_mask: memory allocation flags
2256 int blk_rq_map_kern(request_queue_t
*q
, struct request
*rq
, void *kbuf
,
2257 unsigned int len
, unsigned int gfp_mask
)
2261 if (len
> (q
->max_sectors
<< 9))
2266 bio
= bio_map_kern(q
, kbuf
, len
, gfp_mask
);
2268 return PTR_ERR(bio
);
2270 if (rq_data_dir(rq
) == WRITE
)
2271 bio
->bi_rw
|= (1 << BIO_RW
);
2273 rq
->bio
= rq
->biotail
= bio
;
2274 blk_rq_bio_prep(q
, rq
, bio
);
2276 rq
->buffer
= rq
->data
= NULL
;
2281 EXPORT_SYMBOL(blk_rq_map_kern
);
2284 * blk_execute_rq_nowait - insert a request into queue for execution
2285 * @q: queue to insert the request in
2286 * @bd_disk: matching gendisk
2287 * @rq: request to insert
2288 * @at_head: insert request at head or tail of queue
2289 * @done: I/O completion handler
2292 * Insert a fully prepared request at the back of the io scheduler queue
2293 * for execution. Don't wait for completion.
2295 void blk_execute_rq_nowait(request_queue_t
*q
, struct gendisk
*bd_disk
,
2296 struct request
*rq
, int at_head
,
2297 void (*done
)(struct request
*))
2299 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
2301 rq
->rq_disk
= bd_disk
;
2302 rq
->flags
|= REQ_NOMERGE
;
2304 elv_add_request(q
, rq
, where
, 1);
2305 generic_unplug_device(q
);
2309 * blk_execute_rq - insert a request into queue for execution
2310 * @q: queue to insert the request in
2311 * @bd_disk: matching gendisk
2312 * @rq: request to insert
2313 * @at_head: insert request at head or tail of queue
2316 * Insert a fully prepared request at the back of the io scheduler queue
2317 * for execution and wait for completion.
2319 int blk_execute_rq(request_queue_t
*q
, struct gendisk
*bd_disk
,
2320 struct request
*rq
, int at_head
)
2322 DECLARE_COMPLETION(wait
);
2323 char sense
[SCSI_SENSE_BUFFERSIZE
];
2327 * we need an extra reference to the request, so we can look at
2328 * it after io completion
2333 memset(sense
, 0, sizeof(sense
));
2338 rq
->waiting
= &wait
;
2339 blk_execute_rq_nowait(q
, bd_disk
, rq
, at_head
, blk_end_sync_rq
);
2340 wait_for_completion(&wait
);
2349 EXPORT_SYMBOL(blk_execute_rq
);
2352 * blkdev_issue_flush - queue a flush
2353 * @bdev: blockdev to issue flush for
2354 * @error_sector: error sector
2357 * Issue a flush for the block device in question. Caller can supply
2358 * room for storing the error offset in case of a flush error, if they
2359 * wish to. Caller must run wait_for_completion() on its own.
2361 int blkdev_issue_flush(struct block_device
*bdev
, sector_t
*error_sector
)
2365 if (bdev
->bd_disk
== NULL
)
2368 q
= bdev_get_queue(bdev
);
2371 if (!q
->issue_flush_fn
)
2374 return q
->issue_flush_fn(q
, bdev
->bd_disk
, error_sector
);
2377 EXPORT_SYMBOL(blkdev_issue_flush
);
2379 static void drive_stat_acct(struct request
*rq
, int nr_sectors
, int new_io
)
2381 int rw
= rq_data_dir(rq
);
2383 if (!blk_fs_request(rq
) || !rq
->rq_disk
)
2387 __disk_stat_add(rq
->rq_disk
, read_sectors
, nr_sectors
);
2389 __disk_stat_inc(rq
->rq_disk
, read_merges
);
2390 } else if (rw
== WRITE
) {
2391 __disk_stat_add(rq
->rq_disk
, write_sectors
, nr_sectors
);
2393 __disk_stat_inc(rq
->rq_disk
, write_merges
);
2396 disk_round_stats(rq
->rq_disk
);
2397 rq
->rq_disk
->in_flight
++;
2402 * add-request adds a request to the linked list.
2403 * queue lock is held and interrupts disabled, as we muck with the
2404 * request queue list.
2406 static inline void add_request(request_queue_t
* q
, struct request
* req
)
2408 drive_stat_acct(req
, req
->nr_sectors
, 1);
2411 q
->activity_fn(q
->activity_data
, rq_data_dir(req
));
2414 * elevator indicated where it wants this request to be
2415 * inserted at elevator_merge time
2417 __elv_add_request(q
, req
, ELEVATOR_INSERT_SORT
, 0);
2421 * disk_round_stats() - Round off the performance stats on a struct
2424 * The average IO queue length and utilisation statistics are maintained
2425 * by observing the current state of the queue length and the amount of
2426 * time it has been in this state for.
2428 * Normally, that accounting is done on IO completion, but that can result
2429 * in more than a second's worth of IO being accounted for within any one
2430 * second, leading to >100% utilisation. To deal with that, we call this
2431 * function to do a round-off before returning the results when reading
2432 * /proc/diskstats. This accounts immediately for all queue usage up to
2433 * the current jiffies and restarts the counters again.
2435 void disk_round_stats(struct gendisk
*disk
)
2437 unsigned long now
= jiffies
;
2439 if (now
== disk
->stamp
)
2442 if (disk
->in_flight
) {
2443 __disk_stat_add(disk
, time_in_queue
,
2444 disk
->in_flight
* (now
- disk
->stamp
));
2445 __disk_stat_add(disk
, io_ticks
, (now
- disk
->stamp
));
2451 * queue lock must be held
2453 static void __blk_put_request(request_queue_t
*q
, struct request
*req
)
2455 struct request_list
*rl
= req
->rl
;
2459 if (unlikely(--req
->ref_count
))
2462 elv_completed_request(q
, req
);
2464 req
->rq_status
= RQ_INACTIVE
;
2468 * Request may not have originated from ll_rw_blk. if not,
2469 * it didn't come out of our reserved rq pools
2472 int rw
= rq_data_dir(req
);
2474 BUG_ON(!list_empty(&req
->queuelist
));
2476 blk_free_request(q
, req
);
2477 freed_request(q
, rw
);
2481 void blk_put_request(struct request
*req
)
2483 unsigned long flags
;
2484 request_queue_t
*q
= req
->q
;
2487 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
2488 * following if (q) test.
2491 spin_lock_irqsave(q
->queue_lock
, flags
);
2492 __blk_put_request(q
, req
);
2493 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2497 EXPORT_SYMBOL(blk_put_request
);
2500 * blk_end_sync_rq - executes a completion event on a request
2501 * @rq: request to complete
2503 void blk_end_sync_rq(struct request
*rq
)
2505 struct completion
*waiting
= rq
->waiting
;
2508 __blk_put_request(rq
->q
, rq
);
2511 * complete last, if this is a stack request the process (and thus
2512 * the rq pointer) could be invalid right after this complete()
2516 EXPORT_SYMBOL(blk_end_sync_rq
);
2519 * blk_congestion_wait - wait for a queue to become uncongested
2520 * @rw: READ or WRITE
2521 * @timeout: timeout in jiffies
2523 * Waits for up to @timeout jiffies for a queue (any queue) to exit congestion.
2524 * If no queues are congested then just wait for the next request to be
2527 long blk_congestion_wait(int rw
, long timeout
)
2531 wait_queue_head_t
*wqh
= &congestion_wqh
[rw
];
2533 prepare_to_wait(wqh
, &wait
, TASK_UNINTERRUPTIBLE
);
2534 ret
= io_schedule_timeout(timeout
);
2535 finish_wait(wqh
, &wait
);
2539 EXPORT_SYMBOL(blk_congestion_wait
);
2542 * Has to be called with the request spinlock acquired
2544 static int attempt_merge(request_queue_t
*q
, struct request
*req
,
2545 struct request
*next
)
2547 if (!rq_mergeable(req
) || !rq_mergeable(next
))
2553 if (req
->sector
+ req
->nr_sectors
!= next
->sector
)
2556 if (rq_data_dir(req
) != rq_data_dir(next
)
2557 || req
->rq_disk
!= next
->rq_disk
2558 || next
->waiting
|| next
->special
)
2562 * If we are allowed to merge, then append bio list
2563 * from next to rq and release next. merge_requests_fn
2564 * will have updated segment counts, update sector
2567 if (!q
->merge_requests_fn(q
, req
, next
))
2571 * At this point we have either done a back merge
2572 * or front merge. We need the smaller start_time of
2573 * the merged requests to be the current request
2574 * for accounting purposes.
2576 if (time_after(req
->start_time
, next
->start_time
))
2577 req
->start_time
= next
->start_time
;
2579 req
->biotail
->bi_next
= next
->bio
;
2580 req
->biotail
= next
->biotail
;
2582 req
->nr_sectors
= req
->hard_nr_sectors
+= next
->hard_nr_sectors
;
2584 elv_merge_requests(q
, req
, next
);
2587 disk_round_stats(req
->rq_disk
);
2588 req
->rq_disk
->in_flight
--;
2591 req
->ioprio
= ioprio_best(req
->ioprio
, next
->ioprio
);
2593 __blk_put_request(q
, next
);
2597 static inline int attempt_back_merge(request_queue_t
*q
, struct request
*rq
)
2599 struct request
*next
= elv_latter_request(q
, rq
);
2602 return attempt_merge(q
, rq
, next
);
2607 static inline int attempt_front_merge(request_queue_t
*q
, struct request
*rq
)
2609 struct request
*prev
= elv_former_request(q
, rq
);
2612 return attempt_merge(q
, prev
, rq
);
2618 * blk_attempt_remerge - attempt to remerge active head with next request
2619 * @q: The &request_queue_t belonging to the device
2620 * @rq: The head request (usually)
2623 * For head-active devices, the queue can easily be unplugged so quickly
2624 * that proper merging is not done on the front request. This may hurt
2625 * performance greatly for some devices. The block layer cannot safely
2626 * do merging on that first request for these queues, but the driver can
2627 * call this function and make it happen any way. Only the driver knows
2628 * when it is safe to do so.
2630 void blk_attempt_remerge(request_queue_t
*q
, struct request
*rq
)
2632 unsigned long flags
;
2634 spin_lock_irqsave(q
->queue_lock
, flags
);
2635 attempt_back_merge(q
, rq
);
2636 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2639 EXPORT_SYMBOL(blk_attempt_remerge
);
2641 static int __make_request(request_queue_t
*q
, struct bio
*bio
)
2643 struct request
*req
;
2644 int el_ret
, rw
, nr_sectors
, cur_nr_sectors
, barrier
, err
, sync
;
2645 unsigned short prio
;
2648 sector
= bio
->bi_sector
;
2649 nr_sectors
= bio_sectors(bio
);
2650 cur_nr_sectors
= bio_cur_sectors(bio
);
2651 prio
= bio_prio(bio
);
2653 rw
= bio_data_dir(bio
);
2654 sync
= bio_sync(bio
);
2657 * low level driver can indicate that it wants pages above a
2658 * certain limit bounced to low memory (ie for highmem, or even
2659 * ISA dma in theory)
2661 blk_queue_bounce(q
, &bio
);
2663 spin_lock_prefetch(q
->queue_lock
);
2665 barrier
= bio_barrier(bio
);
2666 if (unlikely(barrier
) && (q
->ordered
== QUEUE_ORDERED_NONE
)) {
2671 spin_lock_irq(q
->queue_lock
);
2673 if (unlikely(barrier
) || elv_queue_empty(q
))
2676 el_ret
= elv_merge(q
, &req
, bio
);
2678 case ELEVATOR_BACK_MERGE
:
2679 BUG_ON(!rq_mergeable(req
));
2681 if (!q
->back_merge_fn(q
, req
, bio
))
2684 req
->biotail
->bi_next
= bio
;
2686 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
2687 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
2688 drive_stat_acct(req
, nr_sectors
, 0);
2689 if (!attempt_back_merge(q
, req
))
2690 elv_merged_request(q
, req
);
2693 case ELEVATOR_FRONT_MERGE
:
2694 BUG_ON(!rq_mergeable(req
));
2696 if (!q
->front_merge_fn(q
, req
, bio
))
2699 bio
->bi_next
= req
->bio
;
2703 * may not be valid. if the low level driver said
2704 * it didn't need a bounce buffer then it better
2705 * not touch req->buffer either...
2707 req
->buffer
= bio_data(bio
);
2708 req
->current_nr_sectors
= cur_nr_sectors
;
2709 req
->hard_cur_sectors
= cur_nr_sectors
;
2710 req
->sector
= req
->hard_sector
= sector
;
2711 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
2712 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
2713 drive_stat_acct(req
, nr_sectors
, 0);
2714 if (!attempt_front_merge(q
, req
))
2715 elv_merged_request(q
, req
);
2718 /* ELV_NO_MERGE: elevator says don't/can't merge. */
2725 * Grab a free request. This is might sleep but can not fail.
2726 * Returns with the queue unlocked.
2728 req
= get_request_wait(q
, rw
, bio
);
2731 * After dropping the lock and possibly sleeping here, our request
2732 * may now be mergeable after it had proven unmergeable (above).
2733 * We don't worry about that case for efficiency. It won't happen
2734 * often, and the elevators are able to handle it.
2737 req
->flags
|= REQ_CMD
;
2740 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2742 if (bio_rw_ahead(bio
) || bio_failfast(bio
))
2743 req
->flags
|= REQ_FAILFAST
;
2746 * REQ_BARRIER implies no merging, but lets make it explicit
2748 if (unlikely(barrier
))
2749 req
->flags
|= (REQ_HARDBARRIER
| REQ_NOMERGE
);
2752 req
->hard_sector
= req
->sector
= sector
;
2753 req
->hard_nr_sectors
= req
->nr_sectors
= nr_sectors
;
2754 req
->current_nr_sectors
= req
->hard_cur_sectors
= cur_nr_sectors
;
2755 req
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2756 req
->nr_hw_segments
= bio_hw_segments(q
, bio
);
2757 req
->buffer
= bio_data(bio
); /* see ->buffer comment above */
2758 req
->waiting
= NULL
;
2759 req
->bio
= req
->biotail
= bio
;
2761 req
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2762 req
->start_time
= jiffies
;
2764 spin_lock_irq(q
->queue_lock
);
2765 if (elv_queue_empty(q
))
2767 add_request(q
, req
);
2770 __generic_unplug_device(q
);
2772 spin_unlock_irq(q
->queue_lock
);
2776 bio_endio(bio
, nr_sectors
<< 9, err
);
2781 * If bio->bi_dev is a partition, remap the location
2783 static inline void blk_partition_remap(struct bio
*bio
)
2785 struct block_device
*bdev
= bio
->bi_bdev
;
2787 if (bdev
!= bdev
->bd_contains
) {
2788 struct hd_struct
*p
= bdev
->bd_part
;
2790 switch (bio_data_dir(bio
)) {
2792 p
->read_sectors
+= bio_sectors(bio
);
2796 p
->write_sectors
+= bio_sectors(bio
);
2800 bio
->bi_sector
+= p
->start_sect
;
2801 bio
->bi_bdev
= bdev
->bd_contains
;
2805 void blk_finish_queue_drain(request_queue_t
*q
)
2807 struct request_list
*rl
= &q
->rq
;
2811 spin_lock_irq(q
->queue_lock
);
2812 clear_bit(QUEUE_FLAG_DRAIN
, &q
->queue_flags
);
2814 while (!list_empty(&q
->drain_list
)) {
2815 rq
= list_entry_rq(q
->drain_list
.next
);
2817 list_del_init(&rq
->queuelist
);
2818 elv_requeue_request(q
, rq
);
2825 spin_unlock_irq(q
->queue_lock
);
2827 wake_up(&rl
->wait
[0]);
2828 wake_up(&rl
->wait
[1]);
2829 wake_up(&rl
->drain
);
2832 static int wait_drain(request_queue_t
*q
, struct request_list
*rl
, int dispatch
)
2834 int wait
= rl
->count
[READ
] + rl
->count
[WRITE
];
2837 wait
+= !list_empty(&q
->queue_head
);
2843 * We rely on the fact that only requests allocated through blk_alloc_request()
2844 * have io scheduler private data structures associated with them. Any other
2845 * type of request (allocated on stack or through kmalloc()) should not go
2846 * to the io scheduler core, but be attached to the queue head instead.
2848 void blk_wait_queue_drained(request_queue_t
*q
, int wait_dispatch
)
2850 struct request_list
*rl
= &q
->rq
;
2853 spin_lock_irq(q
->queue_lock
);
2854 set_bit(QUEUE_FLAG_DRAIN
, &q
->queue_flags
);
2856 while (wait_drain(q
, rl
, wait_dispatch
)) {
2857 prepare_to_wait(&rl
->drain
, &wait
, TASK_UNINTERRUPTIBLE
);
2859 if (wait_drain(q
, rl
, wait_dispatch
)) {
2860 __generic_unplug_device(q
);
2861 spin_unlock_irq(q
->queue_lock
);
2863 spin_lock_irq(q
->queue_lock
);
2866 finish_wait(&rl
->drain
, &wait
);
2869 spin_unlock_irq(q
->queue_lock
);
2873 * block waiting for the io scheduler being started again.
2875 static inline void block_wait_queue_running(request_queue_t
*q
)
2879 while (unlikely(test_bit(QUEUE_FLAG_DRAIN
, &q
->queue_flags
))) {
2880 struct request_list
*rl
= &q
->rq
;
2882 prepare_to_wait_exclusive(&rl
->drain
, &wait
,
2883 TASK_UNINTERRUPTIBLE
);
2886 * re-check the condition. avoids using prepare_to_wait()
2887 * in the fast path (queue is running)
2889 if (test_bit(QUEUE_FLAG_DRAIN
, &q
->queue_flags
))
2892 finish_wait(&rl
->drain
, &wait
);
2896 static void handle_bad_sector(struct bio
*bio
)
2898 char b
[BDEVNAME_SIZE
];
2900 printk(KERN_INFO
"attempt to access beyond end of device\n");
2901 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
2902 bdevname(bio
->bi_bdev
, b
),
2904 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
2905 (long long)(bio
->bi_bdev
->bd_inode
->i_size
>> 9));
2907 set_bit(BIO_EOF
, &bio
->bi_flags
);
2911 * generic_make_request: hand a buffer to its device driver for I/O
2912 * @bio: The bio describing the location in memory and on the device.
2914 * generic_make_request() is used to make I/O requests of block
2915 * devices. It is passed a &struct bio, which describes the I/O that needs
2918 * generic_make_request() does not return any status. The
2919 * success/failure status of the request, along with notification of
2920 * completion, is delivered asynchronously through the bio->bi_end_io
2921 * function described (one day) else where.
2923 * The caller of generic_make_request must make sure that bi_io_vec
2924 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2925 * set to describe the device address, and the
2926 * bi_end_io and optionally bi_private are set to describe how
2927 * completion notification should be signaled.
2929 * generic_make_request and the drivers it calls may use bi_next if this
2930 * bio happens to be merged with someone else, and may change bi_dev and
2931 * bi_sector for remaps as it sees fit. So the values of these fields
2932 * should NOT be depended on after the call to generic_make_request.
2934 void generic_make_request(struct bio
*bio
)
2938 int ret
, nr_sectors
= bio_sectors(bio
);
2941 /* Test device or partition size, when known. */
2942 maxsector
= bio
->bi_bdev
->bd_inode
->i_size
>> 9;
2944 sector_t sector
= bio
->bi_sector
;
2946 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
2948 * This may well happen - the kernel calls bread()
2949 * without checking the size of the device, e.g., when
2950 * mounting a device.
2952 handle_bad_sector(bio
);
2958 * Resolve the mapping until finished. (drivers are
2959 * still free to implement/resolve their own stacking
2960 * by explicitly returning 0)
2962 * NOTE: we don't repeat the blk_size check for each new device.
2963 * Stacking drivers are expected to know what they are doing.
2966 char b
[BDEVNAME_SIZE
];
2968 q
= bdev_get_queue(bio
->bi_bdev
);
2971 "generic_make_request: Trying to access "
2972 "nonexistent block-device %s (%Lu)\n",
2973 bdevname(bio
->bi_bdev
, b
),
2974 (long long) bio
->bi_sector
);
2976 bio_endio(bio
, bio
->bi_size
, -EIO
);
2980 if (unlikely(bio_sectors(bio
) > q
->max_hw_sectors
)) {
2981 printk("bio too big device %s (%u > %u)\n",
2982 bdevname(bio
->bi_bdev
, b
),
2988 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
2991 block_wait_queue_running(q
);
2994 * If this device has partitions, remap block n
2995 * of partition p to block n+start(p) of the disk.
2997 blk_partition_remap(bio
);
2999 ret
= q
->make_request_fn(q
, bio
);
3003 EXPORT_SYMBOL(generic_make_request
);
3006 * submit_bio: submit a bio to the block device layer for I/O
3007 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
3008 * @bio: The &struct bio which describes the I/O
3010 * submit_bio() is very similar in purpose to generic_make_request(), and
3011 * uses that function to do most of the work. Both are fairly rough
3012 * interfaces, @bio must be presetup and ready for I/O.
3015 void submit_bio(int rw
, struct bio
*bio
)
3017 int count
= bio_sectors(bio
);
3019 BIO_BUG_ON(!bio
->bi_size
);
3020 BIO_BUG_ON(!bio
->bi_io_vec
);
3023 mod_page_state(pgpgout
, count
);
3025 mod_page_state(pgpgin
, count
);
3027 if (unlikely(block_dump
)) {
3028 char b
[BDEVNAME_SIZE
];
3029 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s\n",
3030 current
->comm
, current
->pid
,
3031 (rw
& WRITE
) ? "WRITE" : "READ",
3032 (unsigned long long)bio
->bi_sector
,
3033 bdevname(bio
->bi_bdev
,b
));
3036 generic_make_request(bio
);
3039 EXPORT_SYMBOL(submit_bio
);
3041 static void blk_recalc_rq_segments(struct request
*rq
)
3043 struct bio
*bio
, *prevbio
= NULL
;
3044 int nr_phys_segs
, nr_hw_segs
;
3045 unsigned int phys_size
, hw_size
;
3046 request_queue_t
*q
= rq
->q
;
3051 phys_size
= hw_size
= nr_phys_segs
= nr_hw_segs
= 0;
3052 rq_for_each_bio(bio
, rq
) {
3053 /* Force bio hw/phys segs to be recalculated. */
3054 bio
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
3056 nr_phys_segs
+= bio_phys_segments(q
, bio
);
3057 nr_hw_segs
+= bio_hw_segments(q
, bio
);
3059 int pseg
= phys_size
+ prevbio
->bi_size
+ bio
->bi_size
;
3060 int hseg
= hw_size
+ prevbio
->bi_size
+ bio
->bi_size
;
3062 if (blk_phys_contig_segment(q
, prevbio
, bio
) &&
3063 pseg
<= q
->max_segment_size
) {
3065 phys_size
+= prevbio
->bi_size
+ bio
->bi_size
;
3069 if (blk_hw_contig_segment(q
, prevbio
, bio
) &&
3070 hseg
<= q
->max_segment_size
) {
3072 hw_size
+= prevbio
->bi_size
+ bio
->bi_size
;
3079 rq
->nr_phys_segments
= nr_phys_segs
;
3080 rq
->nr_hw_segments
= nr_hw_segs
;
3083 static void blk_recalc_rq_sectors(struct request
*rq
, int nsect
)
3085 if (blk_fs_request(rq
)) {
3086 rq
->hard_sector
+= nsect
;
3087 rq
->hard_nr_sectors
-= nsect
;
3090 * Move the I/O submission pointers ahead if required.
3092 if ((rq
->nr_sectors
>= rq
->hard_nr_sectors
) &&
3093 (rq
->sector
<= rq
->hard_sector
)) {
3094 rq
->sector
= rq
->hard_sector
;
3095 rq
->nr_sectors
= rq
->hard_nr_sectors
;
3096 rq
->hard_cur_sectors
= bio_cur_sectors(rq
->bio
);
3097 rq
->current_nr_sectors
= rq
->hard_cur_sectors
;
3098 rq
->buffer
= bio_data(rq
->bio
);
3102 * if total number of sectors is less than the first segment
3103 * size, something has gone terribly wrong
3105 if (rq
->nr_sectors
< rq
->current_nr_sectors
) {
3106 printk("blk: request botched\n");
3107 rq
->nr_sectors
= rq
->current_nr_sectors
;
3112 static int __end_that_request_first(struct request
*req
, int uptodate
,
3115 int total_bytes
, bio_nbytes
, error
, next_idx
= 0;
3119 * extend uptodate bool to allow < 0 value to be direct io error
3122 if (end_io_error(uptodate
))
3123 error
= !uptodate
? -EIO
: uptodate
;
3126 * for a REQ_BLOCK_PC request, we want to carry any eventual
3127 * sense key with us all the way through
3129 if (!blk_pc_request(req
))
3133 if (blk_fs_request(req
) && !(req
->flags
& REQ_QUIET
))
3134 printk("end_request: I/O error, dev %s, sector %llu\n",
3135 req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
3136 (unsigned long long)req
->sector
);
3139 total_bytes
= bio_nbytes
= 0;
3140 while ((bio
= req
->bio
) != NULL
) {
3143 if (nr_bytes
>= bio
->bi_size
) {
3144 req
->bio
= bio
->bi_next
;
3145 nbytes
= bio
->bi_size
;
3146 bio_endio(bio
, nbytes
, error
);
3150 int idx
= bio
->bi_idx
+ next_idx
;
3152 if (unlikely(bio
->bi_idx
>= bio
->bi_vcnt
)) {
3153 blk_dump_rq_flags(req
, "__end_that");
3154 printk("%s: bio idx %d >= vcnt %d\n",
3156 bio
->bi_idx
, bio
->bi_vcnt
);
3160 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
3161 BIO_BUG_ON(nbytes
> bio
->bi_size
);
3164 * not a complete bvec done
3166 if (unlikely(nbytes
> nr_bytes
)) {
3167 bio_nbytes
+= nr_bytes
;
3168 total_bytes
+= nr_bytes
;
3173 * advance to the next vector
3176 bio_nbytes
+= nbytes
;
3179 total_bytes
+= nbytes
;
3182 if ((bio
= req
->bio
)) {
3184 * end more in this run, or just return 'not-done'
3186 if (unlikely(nr_bytes
<= 0))
3198 * if the request wasn't completed, update state
3201 bio_endio(bio
, bio_nbytes
, error
);
3202 bio
->bi_idx
+= next_idx
;
3203 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
3204 bio_iovec(bio
)->bv_len
-= nr_bytes
;
3207 blk_recalc_rq_sectors(req
, total_bytes
>> 9);
3208 blk_recalc_rq_segments(req
);
3213 * end_that_request_first - end I/O on a request
3214 * @req: the request being processed
3215 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3216 * @nr_sectors: number of sectors to end I/O on
3219 * Ends I/O on a number of sectors attached to @req, and sets it up
3220 * for the next range of segments (if any) in the cluster.
3223 * 0 - we are done with this request, call end_that_request_last()
3224 * 1 - still buffers pending for this request
3226 int end_that_request_first(struct request
*req
, int uptodate
, int nr_sectors
)
3228 return __end_that_request_first(req
, uptodate
, nr_sectors
<< 9);
3231 EXPORT_SYMBOL(end_that_request_first
);
3234 * end_that_request_chunk - end I/O on a request
3235 * @req: the request being processed
3236 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3237 * @nr_bytes: number of bytes to complete
3240 * Ends I/O on a number of bytes attached to @req, and sets it up
3241 * for the next range of segments (if any). Like end_that_request_first(),
3242 * but deals with bytes instead of sectors.
3245 * 0 - we are done with this request, call end_that_request_last()
3246 * 1 - still buffers pending for this request
3248 int end_that_request_chunk(struct request
*req
, int uptodate
, int nr_bytes
)
3250 return __end_that_request_first(req
, uptodate
, nr_bytes
);
3253 EXPORT_SYMBOL(end_that_request_chunk
);
3256 * queue lock must be held
3258 void end_that_request_last(struct request
*req
)
3260 struct gendisk
*disk
= req
->rq_disk
;
3262 if (unlikely(laptop_mode
) && blk_fs_request(req
))
3263 laptop_io_completion();
3265 if (disk
&& blk_fs_request(req
)) {
3266 unsigned long duration
= jiffies
- req
->start_time
;
3267 switch (rq_data_dir(req
)) {
3269 __disk_stat_inc(disk
, writes
);
3270 __disk_stat_add(disk
, write_ticks
, duration
);
3273 __disk_stat_inc(disk
, reads
);
3274 __disk_stat_add(disk
, read_ticks
, duration
);
3277 disk_round_stats(disk
);
3283 __blk_put_request(req
->q
, req
);
3286 EXPORT_SYMBOL(end_that_request_last
);
3288 void end_request(struct request
*req
, int uptodate
)
3290 if (!end_that_request_first(req
, uptodate
, req
->hard_cur_sectors
)) {
3291 add_disk_randomness(req
->rq_disk
);
3292 blkdev_dequeue_request(req
);
3293 end_that_request_last(req
);
3297 EXPORT_SYMBOL(end_request
);
3299 void blk_rq_bio_prep(request_queue_t
*q
, struct request
*rq
, struct bio
*bio
)
3301 /* first three bits are identical in rq->flags and bio->bi_rw */
3302 rq
->flags
|= (bio
->bi_rw
& 7);
3304 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
3305 rq
->nr_hw_segments
= bio_hw_segments(q
, bio
);
3306 rq
->current_nr_sectors
= bio_cur_sectors(bio
);
3307 rq
->hard_cur_sectors
= rq
->current_nr_sectors
;
3308 rq
->hard_nr_sectors
= rq
->nr_sectors
= bio_sectors(bio
);
3309 rq
->buffer
= bio_data(bio
);
3311 rq
->bio
= rq
->biotail
= bio
;
3314 EXPORT_SYMBOL(blk_rq_bio_prep
);
3316 int kblockd_schedule_work(struct work_struct
*work
)
3318 return queue_work(kblockd_workqueue
, work
);
3321 EXPORT_SYMBOL(kblockd_schedule_work
);
3323 void kblockd_flush(void)
3325 flush_workqueue(kblockd_workqueue
);
3327 EXPORT_SYMBOL(kblockd_flush
);
3329 int __init
blk_dev_init(void)
3331 kblockd_workqueue
= create_workqueue("kblockd");
3332 if (!kblockd_workqueue
)
3333 panic("Failed to create kblockd\n");
3335 request_cachep
= kmem_cache_create("blkdev_requests",
3336 sizeof(struct request
), 0, SLAB_PANIC
, NULL
, NULL
);
3338 requestq_cachep
= kmem_cache_create("blkdev_queue",
3339 sizeof(request_queue_t
), 0, SLAB_PANIC
, NULL
, NULL
);
3341 iocontext_cachep
= kmem_cache_create("blkdev_ioc",
3342 sizeof(struct io_context
), 0, SLAB_PANIC
, NULL
, NULL
);
3344 blk_max_low_pfn
= max_low_pfn
;
3345 blk_max_pfn
= max_pfn
;
3351 * IO Context helper functions
3353 void put_io_context(struct io_context
*ioc
)
3358 BUG_ON(atomic_read(&ioc
->refcount
) == 0);
3360 if (atomic_dec_and_test(&ioc
->refcount
)) {
3361 if (ioc
->aic
&& ioc
->aic
->dtor
)
3362 ioc
->aic
->dtor(ioc
->aic
);
3363 if (ioc
->cic
&& ioc
->cic
->dtor
)
3364 ioc
->cic
->dtor(ioc
->cic
);
3366 kmem_cache_free(iocontext_cachep
, ioc
);
3369 EXPORT_SYMBOL(put_io_context
);
3371 /* Called by the exitting task */
3372 void exit_io_context(void)
3374 unsigned long flags
;
3375 struct io_context
*ioc
;
3377 local_irq_save(flags
);
3379 ioc
= current
->io_context
;
3380 current
->io_context
= NULL
;
3382 task_unlock(current
);
3383 local_irq_restore(flags
);
3385 if (ioc
->aic
&& ioc
->aic
->exit
)
3386 ioc
->aic
->exit(ioc
->aic
);
3387 if (ioc
->cic
&& ioc
->cic
->exit
)
3388 ioc
->cic
->exit(ioc
->cic
);
3390 put_io_context(ioc
);
3394 * If the current task has no IO context then create one and initialise it.
3395 * Otherwise, return its existing IO context.
3397 * This returned IO context doesn't have a specifically elevated refcount,
3398 * but since the current task itself holds a reference, the context can be
3399 * used in general code, so long as it stays within `current` context.
3401 struct io_context
*current_io_context(int gfp_flags
)
3403 struct task_struct
*tsk
= current
;
3404 struct io_context
*ret
;
3406 ret
= tsk
->io_context
;
3410 ret
= kmem_cache_alloc(iocontext_cachep
, gfp_flags
);
3412 atomic_set(&ret
->refcount
, 1);
3413 ret
->task
= current
;
3414 ret
->set_ioprio
= NULL
;
3415 ret
->last_waited
= jiffies
; /* doesn't matter... */
3416 ret
->nr_batch_requests
= 0; /* because this is 0 */
3419 tsk
->io_context
= ret
;
3424 EXPORT_SYMBOL(current_io_context
);
3427 * If the current task has no IO context then create one and initialise it.
3428 * If it does have a context, take a ref on it.
3430 * This is always called in the context of the task which submitted the I/O.
3432 struct io_context
*get_io_context(int gfp_flags
)
3434 struct io_context
*ret
;
3435 ret
= current_io_context(gfp_flags
);
3437 atomic_inc(&ret
->refcount
);
3440 EXPORT_SYMBOL(get_io_context
);
3442 void copy_io_context(struct io_context
**pdst
, struct io_context
**psrc
)
3444 struct io_context
*src
= *psrc
;
3445 struct io_context
*dst
= *pdst
;
3448 BUG_ON(atomic_read(&src
->refcount
) == 0);
3449 atomic_inc(&src
->refcount
);
3450 put_io_context(dst
);
3454 EXPORT_SYMBOL(copy_io_context
);
3456 void swap_io_context(struct io_context
**ioc1
, struct io_context
**ioc2
)
3458 struct io_context
*temp
;
3463 EXPORT_SYMBOL(swap_io_context
);
3468 struct queue_sysfs_entry
{
3469 struct attribute attr
;
3470 ssize_t (*show
)(struct request_queue
*, char *);
3471 ssize_t (*store
)(struct request_queue
*, const char *, size_t);
3475 queue_var_show(unsigned int var
, char *page
)
3477 return sprintf(page
, "%d\n", var
);
3481 queue_var_store(unsigned long *var
, const char *page
, size_t count
)
3483 char *p
= (char *) page
;
3485 *var
= simple_strtoul(p
, &p
, 10);
3489 static ssize_t
queue_requests_show(struct request_queue
*q
, char *page
)
3491 return queue_var_show(q
->nr_requests
, (page
));
3495 queue_requests_store(struct request_queue
*q
, const char *page
, size_t count
)
3497 struct request_list
*rl
= &q
->rq
;
3499 int ret
= queue_var_store(&q
->nr_requests
, page
, count
);
3500 if (q
->nr_requests
< BLKDEV_MIN_RQ
)
3501 q
->nr_requests
= BLKDEV_MIN_RQ
;
3502 blk_queue_congestion_threshold(q
);
3504 if (rl
->count
[READ
] >= queue_congestion_on_threshold(q
))
3505 set_queue_congested(q
, READ
);
3506 else if (rl
->count
[READ
] < queue_congestion_off_threshold(q
))
3507 clear_queue_congested(q
, READ
);
3509 if (rl
->count
[WRITE
] >= queue_congestion_on_threshold(q
))
3510 set_queue_congested(q
, WRITE
);
3511 else if (rl
->count
[WRITE
] < queue_congestion_off_threshold(q
))
3512 clear_queue_congested(q
, WRITE
);
3514 if (rl
->count
[READ
] >= q
->nr_requests
) {
3515 blk_set_queue_full(q
, READ
);
3516 } else if (rl
->count
[READ
]+1 <= q
->nr_requests
) {
3517 blk_clear_queue_full(q
, READ
);
3518 wake_up(&rl
->wait
[READ
]);
3521 if (rl
->count
[WRITE
] >= q
->nr_requests
) {
3522 blk_set_queue_full(q
, WRITE
);
3523 } else if (rl
->count
[WRITE
]+1 <= q
->nr_requests
) {
3524 blk_clear_queue_full(q
, WRITE
);
3525 wake_up(&rl
->wait
[WRITE
]);
3530 static ssize_t
queue_ra_show(struct request_queue
*q
, char *page
)
3532 int ra_kb
= q
->backing_dev_info
.ra_pages
<< (PAGE_CACHE_SHIFT
- 10);
3534 return queue_var_show(ra_kb
, (page
));
3538 queue_ra_store(struct request_queue
*q
, const char *page
, size_t count
)
3540 unsigned long ra_kb
;
3541 ssize_t ret
= queue_var_store(&ra_kb
, page
, count
);
3543 spin_lock_irq(q
->queue_lock
);
3544 if (ra_kb
> (q
->max_sectors
>> 1))
3545 ra_kb
= (q
->max_sectors
>> 1);
3547 q
->backing_dev_info
.ra_pages
= ra_kb
>> (PAGE_CACHE_SHIFT
- 10);
3548 spin_unlock_irq(q
->queue_lock
);
3553 static ssize_t
queue_max_sectors_show(struct request_queue
*q
, char *page
)
3555 int max_sectors_kb
= q
->max_sectors
>> 1;
3557 return queue_var_show(max_sectors_kb
, (page
));
3561 queue_max_sectors_store(struct request_queue
*q
, const char *page
, size_t count
)
3563 unsigned long max_sectors_kb
,
3564 max_hw_sectors_kb
= q
->max_hw_sectors
>> 1,
3565 page_kb
= 1 << (PAGE_CACHE_SHIFT
- 10);
3566 ssize_t ret
= queue_var_store(&max_sectors_kb
, page
, count
);
3569 if (max_sectors_kb
> max_hw_sectors_kb
|| max_sectors_kb
< page_kb
)
3572 * Take the queue lock to update the readahead and max_sectors
3573 * values synchronously:
3575 spin_lock_irq(q
->queue_lock
);
3577 * Trim readahead window as well, if necessary:
3579 ra_kb
= q
->backing_dev_info
.ra_pages
<< (PAGE_CACHE_SHIFT
- 10);
3580 if (ra_kb
> max_sectors_kb
)
3581 q
->backing_dev_info
.ra_pages
=
3582 max_sectors_kb
>> (PAGE_CACHE_SHIFT
- 10);
3584 q
->max_sectors
= max_sectors_kb
<< 1;
3585 spin_unlock_irq(q
->queue_lock
);
3590 static ssize_t
queue_max_hw_sectors_show(struct request_queue
*q
, char *page
)
3592 int max_hw_sectors_kb
= q
->max_hw_sectors
>> 1;
3594 return queue_var_show(max_hw_sectors_kb
, (page
));
3598 static struct queue_sysfs_entry queue_requests_entry
= {
3599 .attr
= {.name
= "nr_requests", .mode
= S_IRUGO
| S_IWUSR
},
3600 .show
= queue_requests_show
,
3601 .store
= queue_requests_store
,
3604 static struct queue_sysfs_entry queue_ra_entry
= {
3605 .attr
= {.name
= "read_ahead_kb", .mode
= S_IRUGO
| S_IWUSR
},
3606 .show
= queue_ra_show
,
3607 .store
= queue_ra_store
,
3610 static struct queue_sysfs_entry queue_max_sectors_entry
= {
3611 .attr
= {.name
= "max_sectors_kb", .mode
= S_IRUGO
| S_IWUSR
},
3612 .show
= queue_max_sectors_show
,
3613 .store
= queue_max_sectors_store
,
3616 static struct queue_sysfs_entry queue_max_hw_sectors_entry
= {
3617 .attr
= {.name
= "max_hw_sectors_kb", .mode
= S_IRUGO
},
3618 .show
= queue_max_hw_sectors_show
,
3621 static struct queue_sysfs_entry queue_iosched_entry
= {
3622 .attr
= {.name
= "scheduler", .mode
= S_IRUGO
| S_IWUSR
},
3623 .show
= elv_iosched_show
,
3624 .store
= elv_iosched_store
,
3627 static struct attribute
*default_attrs
[] = {
3628 &queue_requests_entry
.attr
,
3629 &queue_ra_entry
.attr
,
3630 &queue_max_hw_sectors_entry
.attr
,
3631 &queue_max_sectors_entry
.attr
,
3632 &queue_iosched_entry
.attr
,
3636 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
3639 queue_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
3641 struct queue_sysfs_entry
*entry
= to_queue(attr
);
3642 struct request_queue
*q
;
3644 q
= container_of(kobj
, struct request_queue
, kobj
);
3648 return entry
->show(q
, page
);
3652 queue_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
3653 const char *page
, size_t length
)
3655 struct queue_sysfs_entry
*entry
= to_queue(attr
);
3656 struct request_queue
*q
;
3658 q
= container_of(kobj
, struct request_queue
, kobj
);
3662 return entry
->store(q
, page
, length
);
3665 static struct sysfs_ops queue_sysfs_ops
= {
3666 .show
= queue_attr_show
,
3667 .store
= queue_attr_store
,
3670 static struct kobj_type queue_ktype
= {
3671 .sysfs_ops
= &queue_sysfs_ops
,
3672 .default_attrs
= default_attrs
,
3675 int blk_register_queue(struct gendisk
*disk
)
3679 request_queue_t
*q
= disk
->queue
;
3681 if (!q
|| !q
->request_fn
)
3684 q
->kobj
.parent
= kobject_get(&disk
->kobj
);
3685 if (!q
->kobj
.parent
)
3688 snprintf(q
->kobj
.name
, KOBJ_NAME_LEN
, "%s", "queue");
3689 q
->kobj
.ktype
= &queue_ktype
;
3691 ret
= kobject_register(&q
->kobj
);
3695 ret
= elv_register_queue(q
);
3697 kobject_unregister(&q
->kobj
);
3704 void blk_unregister_queue(struct gendisk
*disk
)
3706 request_queue_t
*q
= disk
->queue
;
3708 if (q
&& q
->request_fn
) {
3709 elv_unregister_queue(q
);
3711 kobject_unregister(&q
->kobj
);
3712 kobject_put(&disk
->kobj
);